2015 Barabasi network course book - book_chapter_2

CID Name MustBeSatisfieCategorySubclausePageLineTopicIssueID2001Hunter,DavidYes Editorial22647GeneralG-Editor2002Hunter,DavidYes Editorial2251GeneralG-Editor2003Hunter,DavidYes Editorial22059SecurityS-Edit2004Hunter,DavidYes Editorial22055GeneralG-Editor2005Hunter,DavidYes Editorial22049SecurityS-General2006Hunter,DavidYes Editorial21942SecurityS-Edit2007Hunter,DavidYes Editorial21924GeneralG-Editor2008Hunter,DavidYes Editorial22014SecurityS-Edit2009Hunter,DavidYes Editorial22012GeneralG-Editor2010Hunter,DavidYes Editorial21965GeneralG-Editor2011Hunter,DavidYes Editorial21958SecurityS-Edit2012Hunter,DavidYes Editorial21954SecurityS-Edit2013Hunter,DavidYes Editorial21952SecurityS-Edit2014Hunter,DavidYes Editorial21949SecurityS-Edit2015Hunter,DavidYes Editorial21936GeneralG-Editor2016Hunter,DavidYes Editorial21934SecurityS-Edit2017Hunter,DavidYes Editorial21924SecurityS-Edit2018Hunter,DavidYes Editorial21921SecurityS-Edit2019Hunter,DavidYes Editorial21918SecurityS-Edit2020Hunter,DavidYes Editorial21916SecurityS-Edit2021Hunter,DavidYes Editorial21913SecurityS-Edit2022Hunter,DavidYes Editorial21912SecurityS-Edit2023Hunter,DavidYes Editorial21911SecurityS-Edit2024Hunter,DavidYes Editorial2198SecurityS-Edit2025Hunter,DavidYes Editorial2196SecurityS-Edit2026Hunter,DavidYes Editorial2191SecurityS-Edit2027Hunter,DavidYes Editorial21863SecurityS-Edit2028Hunter,DavidYes Editorial21861SecurityS-Edit2029Hunter,DavidYes Editorial21860SecurityS-Edit2030Hunter,DavidYes Editorial21829GeneralG-Editor2031Hunter,DavidYes Editorial21826SecurityS-Edit2032Hunter,DavidYes Editorial21826GeneralG-Editor2033Hunter,DavidYes Technical2179SecurityS-MPM2034Hunter,DavidYes Editorial2171GeneralG-Editor2035Hunter,DavidYes Technical20943SecurityS-Edit2036Hunter,DavidYes Editorial20722SecurityS-Edit2037Hunter,DavidYes Technical20718SecurityS-Edit2038Hunter,DavidYes Editorial20718SecurityS-Edit2039Hunter,DavidYes Technical20338SecurityS-Edit2040Hunter,DavidYes Technical20335SecurityS-Edit2041Hunter,DavidYes Editorial17639GeneralG-Prim2042Hunter,DavidYes Editorial17639GeneralG-Prim2043Hunter,DavidYes Editorial17639GeneralG-Prim2044Hunter,DavidYes Editorial17639GeneralG-Prim2045Hunter,DavidYes Editorial17639GeneralG-Prim2046Hunter,DavidYes Editorial17414GeneralG-Prim2047Hunter,DavidYes Technical16321MACM-MCCA2048Hunter,DavidYes Editorial15853GeneralG-Prim2049Hunter,DavidYes Editorial15724GeneralG-Prim2050Hunter,DavidYes Editorial1561GeneralG-Prim2051Hunter,DavidYes Editorial15439GeneralG-Prim2052Hunter,DavidYes Editorial1538GeneralG-Prim2053Hunter,DavidYes Editorial1531GeneralG-Prim2054Hunter,DavidYes Editorial15224GeneralG-Prim2055Hunter,DavidYes Editorial15222GeneralG-Prim2056Hunter,DavidYes Editorial15054GeneralG-Prim2057Hunter,DavidYes Technical14964MAC M-PM2058Hunter,DavidYes Editorial14958GeneralG-Prim2059Hunter,DavidYes Editorial14950GeneralG-Prim2060Hunter,DavidYes Editorial14950GeneralG-Prim2061Hunter,DavidYes Editorial14336GeneralG-Editor2062Hunter,DavidYes Technical14335GeneralG-Editor2063Hunter,DavidYes Technical14332GeneralG-Editor2064Hunter,DavidYes Technical14327GeneralG-Editor2065Hunter,DavidYes Technical13836GeneralG-Editor2066Hunter,DavidYes Editorial13624GeneralG-Editor2067Hunter,DavidYes Editorial11239MACM-MCCA2068Hunter,DavidYes Technical10556SecurityS-Edit2069Hunter,DavidYes Technical9242SecurityS-SAE2070Hunter,DavidYes Technical9046SecurityS-SAE2071Hunter,DavidYes Technical9043SecurityS-SAE2072Hunter,DavidYes Technical909SecurityS-SAE2073Hunter,DavidYes Technical8940SecurityS-SAE2074Hunter,DavidYes Technical8740SecurityS-SAE2075Hunter,DavidYes Technical8722SecurityS-SAE2076Hunter,DavidYes Technical8713SecurityS-SAE2077Hunter,DavidYes Technical8655SecurityS-SAE2078Hunter,DavidYes Technical868SecurityS-SAE2079Hunter,DavidYes Technical868SecurityS-SAE2080Hunter,DavidYes Technical8545SecurityS-SAE2081Hunter,DavidYes Technical8544SecurityS-SAE2082Hunter,DavidYes Technical6528RFIR-Proxy2083Hunter,DavidYes Technical6161RFIR-HWMP2084Hunter,DavidYes Technical5548MACM-MCCA2085Hunter,DavidYes Technical3643SecurityS-SAE2086Hunter,DavidYes Technical3628SecurityS-SAE2087Hunter,DavidYes Technical3611SecurityS-SAE2088Hunter,DavidYes Editorial3611SecurityS-SAE2089Hunter,DavidYes Technical3549SecurityS-SAE2090Hunter,DavidYes Editorial3423GeneralG-Editor2091Hunter,DavidYes Technical1536SecurityS-Edit2092Hunter,DavidYes Technical1316GeneralG-Emergency2093Hunter,DavidYes Technical1321GeneralG-Emergency2094Hunter,DavidYes Technical1264RFIR-Proxy2095Hunter,DavidYes Editorial1260GeneralG-Editor2096Hunter,DavidYes Technical1254GeneralG-Editor2097Hunter,DavidYes Editorial126GeneralG-Editor2098Hunter,DavidYes Editorial1131GeneralG-Editor2099Hunter,DavidYes Editorial1130GeneralG-Editor2100Hunter,DavidYes Editorial1129GeneralG-Editor2101Hunter,DavidYes Editorial1129GeneralG-Editor2102Hunter,DavidYes Editorial1123GeneralG-Editor2103Hunter,DavidNo Editorial118GeneralG-Editor2104Hunter,DavidYes Editorial1057GeneralG-Editor2105Hunter,DavidYes Editorial1042GeneralG-Editor2106Hunter,DavidYes Technical102GeneralG-Editor2107Hunter,DavidYes Editorial865GeneralG-Arch2108Hunter,DavidYes Editorial859GeneralG-Arch2109Hunter,DavidYes Technical859GeneralG-Arch2110Hunter,DavidYes Technical856GeneralG-Arch2111Hunter,DavidYes Editorial856GeneralG-Editor2112Hunter,DavidYes Technical849GeneralG-Base2113Hunter,DavidYes Editorial838GeneralG-Base2114Hunter,DavidYes Editorial836GeneralG-Base2115Hunter,DavidYes Editorial830GeneralG-Arch2116Hunter,DavidYes Editorial828GeneralG-Arch2117Hunter,DavidYes Editorial823GeneralG-Arch2118Hunter,DavidYes Technical816GeneralG-Arch2119Hunter,DavidYes Editorial816GeneralG-Editor2120Hunter,DavidYes Technical815GeneralG-Editor2121Hunter,DavidYes Technical87GeneralG-Editor2122Hunter,DavidYes Technical81GeneralG-Arch2123Hunter,DavidYes Editorial81GeneralG-Editor2124Hunter,DavidYes Editorial764GeneralG-Editor2125Hunter,DavidYes Editorial739GeneralG-Arch2126Hunter,DavidYes Technical733GeneralG-Arch2127Hunter,DavidYes Technical730GeneralG-Arch2128Hunter,DavidYes Technical716GeneralG-Arch2129Hunter,DavidNo Editorial542GeneralG-Editor2130Hunter,DavidYes Technical528MACM-MCCA2131Hunter,DavidYes Editorial515MAC M-PM2132Hunter,DavidYes Editorial51RFIR-General2133Hunter,DavidYes Technical45GeneralG-Arch2134Hunter,DavidYes Technical349RFIR-MeshGate2135Hunter,DavidYes Editorial318GeneralG-Editor2136Bahr,MichaelYes Technical336GeneralG-Base2137Bahr,MichaelYes Technical7.1.3.1.71929MAC M-PM2138Bahr,MichaelYes Technical7.1.3.1.71923MAC M-PM2139Bahr,MichaelYes Technical7.1.3.6.32337GeneralG-Frame2140Bahr,MichaelYes Technical7.3.2379GeneralG-Frame2141Bahr,MichaelNo Technical7.3.23761GeneralG-Editor2142Bahr,MichaelYes Technical7.3.23761MACM-MCCA2143Bahr,MichaelYes Technical7.3.23739MACM-MCCA2144Bahr,MichaelYes Technical7.3.23737MACM-MCCA2145Bahr,MichaelNo Editorial7.3.2.904145GeneralG-Emergency2146Bahr,MichaelYes Technical7.3.2.98.74547GeneralG-Frame2147Bahr,MichaelYes Technical7.3.2.98.94655GeneralG-General2148Bahr,MichaelYes Technical7.3.2.1055221MAC M-BS2149Bahr,MichaelNo Editorial7.3.2.1085464MACM-MCCA2150Bahr,MichaelNo Editorial7.3.2.1085440MACM-MCCA2151Bahr,MichaelNo Editorial7.3.2.1085440MACM-MCCA2152Bahr,MichaelNo Editorial7.3.2.1085536MACM-MCCA2153Bahr,MichaelNo Editorial7.3.2.1085526MACM-MCCA2154Bahr,MichaelNo Editorial7.3.2.1085510MACM-MCCA2155Bahr,MichaelNo Technical7.3.2.109.25664MACM-MCCA2156Bahr,MichaelNo Technical7.3.2.109.25660MACM-MCCA2157Bahr,MichaelNo Editorial7.3.2.109.15632MACM-MCCA2158Bahr,MichaelNo Editorial7.3.2.109.15623MACM-MCCA2159Bahr,MichaelNo Editorial7.3.2.109.15616MACM-MCCA2160Bahr,MichaelNo Editorial7.3.2.109.35728MACM-MCCA2161Bahr,MichaelNo Technical7.3.2.109.35725MACM-MCCA2162Bahr,MichaelNo Technical7.3.2.109.2571MACM-MCCA2163Bahr,MichaelNo Editorial7.3.2.1136047RFIR-MeshGate2164Bahr,MichaelNo Technical7.3.2.1166513RFIR-Proxy2165Bahr,MichaelNo Editorial7.4.15784MACM-MCCA2166Bahr,MichaelNo Editorial7.4.15.97958MACM-MCCA2167Bahr,MichaelNo Editorial9.9a.3.211343MACM-MCCA2168Bahr,MichaelYes Technical9.9a.3.111334MACM-MCCA2169Bahr,MichaelNo Technical9.9a.3.111339MACM-MCCA2170Bahr,MichaelNo Editorial9.9a.3.111315MACM-MCCA2171Bahr,MichaelNo Editorial9.9a.3.311416MACM-MCCA2172Bahr,MichaelNo Editorial9.9a.3.7.4b12060MACM-MCCA2173Bahr,MichaelNo Editorial9.9a.3.7.4a12040MACM-MCCA2174Bahr,MichaelNo Editorial9.9a.3.7.412022MACM-MCCA2175Bahr,MichaelNo Editorial9.9a.3.7.4b1215MACM-MCCA2176Bahr,MichaelNo Editorial9.9a.3.1012427MACM-MCCA2177Bahr,MichaelNo Technical9.22.212558RFI R-FWD2178Bahr,MichaelNo Technical9.22.212616RFI R-FWD2179Bahr,MichaelNo Editorial9.22.312754RFI R-FWD2180Bahr,MichaelNo Technical10.3.81.1.216318MACM-MCCA2181Bahr,MichaelNo Editorial10.3.81.7.116852MACM-MCCA2182Bahr,MichaelNo Editorial10.3.81.8.216959MACM-MCCA2183Bahr,MichaelNo Editorial10.3.81.8.216953MACM-MCCA2184Bahr,MichaelNo Editorial10.3.81.7.316918MACM-MCCA2185Bahr,MichaelNo Editorial10.3.81.8.216941MACM-MCCA2186Bahr,MichaelNo Editorial10.3.81.10.217127MACM-MCCA2187Bahr,MichaelNo Editorial10.3.81.10.217115MACM-MCCA2188Bahr,MichaelYes Technical11.119034MAC M-BS2189Bahr,MichaelYes Technical11.1.319035MAC M-BS2190Bahr,MichaelYes Technical11.119034MAC M-BS2191Bahr,MichaelNo Editorial11.1.1.319026MAC M-BS2192Bahr,MichaelNo Technical11.1.1.319029MAC M-BS2193Bahr,MichaelNo Editorial11C.9.723762GeneralG-Editor2194Bahr,MichaelYes Technical11C.9.8.323961RFIR-HWMP2195Bahr,MichaelYes Technical11C.9.9.324249RFIR-HWMP2196Bahr,MichaelYes Technical11C.9.9.324357RFIR-HWMP2197Bahr,MichaelYes Technical11C.9.9.324615RFIR-HWMP2198Bahr,MichaelYes Technical11C.9.9.4.325119RFIR-HWMP2199Bahr,MichaelNo Technical11C.9.11.325746RFIR-HWMP2200Bahr,MichaelNo Technical11C.9.11.325952RFIR-HWMP2201Bahr,MichaelYes Technical11C.9.11.4.326044RFIR-HWMP2202Bahr,MichaelYes Technical11C.9.11.4.326038RFIR-HWMP2203Bahr,MichaelYes Technical11C.9.11.4.326033RFIR-HWMP2204Bahr,MichaelNo Technical11C.10.4.326910RFIR-Proxy2205Bahr,MichaelNo Technical11C.12.2.2.127340MAC M-BS2206Bahr,MichaelYes Technical11C.12.2.2.127337MAC M-BS2207Bahr,MichaelNo Technical11C.12.2.2.227354MAC M-BS2208Bahr,MichaelNo Technical11C.12.2.127311MAC M-BS2209Bahr,MichaelNo Technical11C.12.2.227322MAC M-BS2210Bahr,MichaelNo Editorial11C.12.2.2.227365MAC M-BS2211Bahr,MichaelNo Technical11C.12.2.2.127332MAC M-BS2212Bahr,MichaelYes Technical11C.12.2.2.327424MAC M-BS2213Bahr,MichaelNo Technical11C.12.2.2.22741MAC M-BS2214Bahr,MichaelNo Technical11C.12.2.2.327463MAC M-BS2215Bahr,MichaelYes Technical11C.12.2.2.327456MAC M-BS2216Bahr,MichaelYes Technical11C.12.2.2.327457MAC M-BS2217Bahr,MichaelYes Technical11C.12.2.2.327437MAC M-BS2218Bahr,MichaelNo Technical11C.12.2.2.327424MAC M-BS2219Bahr,MichaelYes Technical11C.12.2.2.327435MAC M-BS2220Bahr,MichaelYes Technical11C.12.2.2.327456MAC M-BS2221Bahr,MichaelYes Technical11C.12.2.2.327446MAC M-BS2222Bahr,MichaelNo Technical11C.12.2.2.327430MAC M-BS2223Bahr,MichaelNo Technical11C.12.2.2.227417MAC M-BS2224Bahr,MichaelNo Editorial11C.12.2.2.22746MAC M-BS2225Bahr,MichaelNo Technical11C.12.3.127513MAC M-BS2226Bahr,MichaelNo Technical11C.12.3.227530MAC M-BS2227Bahr,MichaelNo Editorial11C.12.4.127564MAC M-BS2228Bahr,MichaelNo Technical11C.12.4.127562MAC M-BS2229Bahr,MichaelNo Editorial11C.12.3.227530MAC M-BS2230Bahr,MichaelNo General11C.12.3.227540MAC M-BS2231Bahr,MichaelNo Editorial11C.12.3.227534MAC M-BS2232Bahr,MichaelYes Technical11C.12.4.2.327665MAC M-BS2233Bahr,MichaelNo Editorial11C.12.4.2.327663MAC M-BS2234Bahr,MichaelYes Technical11C.12.4.2.227654MAC M-BS2235Bahr,MichaelNo Technical11C.12.4.2.227652MAC M-BS2236Bahr,MichaelNo Editorial11C.12.4.2.227640MAC M-BS2237Bahr,MichaelNo Editorial11C.12.4.2.227634MAC M-BS2238Bahr,MichaelNo Technical11C.12.4.2.127611MAC M-BS2239Bahr,MichaelNo Technical11C.12.4.2.12768MAC M-BS2240Bahr,MichaelNo Technical11C.12.4.2.12767MAC M-BS2241Bahr,MichaelYes Technical11C.12.4.22761MAC M-BS2242Bahr,MichaelNo Editorial11C.12.4.2.427755MAC M-BS2243Bahr,MichaelNo Editorial11C.12.4.2.427749MAC M-BS2244Bahr,MichaelNo Editorial11C.12.4.2.427748MAC M-BS2245Bahr,MichaelNo Editorial11C.12.4.2.427735MAC M-BS2246Bahr,MichaelNo Editorial11C.12.4.2.427734MAC M-BS2247Bahr,MichaelNo Editorial11C.12.4.2.427734MAC M-BS2248Bahr,MichaelNo Editorial11C.12.4.2.427732MAC M-BS2249Bahr,MichaelNo Editorial11C.12.4.2.32771MAC M-BS2250Bahr,MichaelNo Technical11C.12.4.2.32771MAC M-BS2251Bahr,MichaelNo Editorial11C.12.4.327864MAC M-BS2252Bahr,MichaelNo Technical11C.12.4.327856MAC M-BS2253Bahr,MichaelNo Technical11C.12.4.2.527817MAC M-BS2254Bahr,MichaelYes Technical11C.12.4.2.527815MAC M-BS2255Bahr,MichaelNo Editorial11C.12.4.2.527848MAC M-BS2256Bahr,MichaelNo Technical11C.12.4.2.527843MAC M-BS2257Bahr,MichaelNo Editorial11C.12.4.2.527829MAC M-BS2258Bahr,MichaelNo Editorial11C.12.4.2.527821MAC M-BS2259Bahr,MichaelNo Editorial11C.12.4.2.42781MAC M-BS2260Bahr,MichaelYes Technical11C.12.4.4.127923MAC M-BS2261Bahr,MichaelNo Technical11C.12.4.4.127922MAC M-BS2262Bahr,MichaelYes Technical11C.12.4.4.127915MAC M-BS2263Bahr,MichaelNo Editorial11C.12.4.4.127915MAC M-BS2264Bahr,MichaelYes Technical11C.12.4.32792MAC M-BS2265Bahr,MichaelNo Editorial11C.12.4.4.327953MAC M-BS2266Bahr,MichaelYes Technical11C.12.4.4.327962MAC M-BS2267Bahr,MichaelYes Technical11C.12.4.4.327958MAC M-BS2268Bahr,MichaelNo Technical D31310MACM-MCCA2269Bahr,MichaelNo Technical Y.733549RFIR-HWMP2270Strutt,GuenaelNo Technical11C.4.320814SecurityS-MPM2271Strutt,GuenaelNo Technical11C.4.220644SecurityS-MPM2272Strutt,GuenaelNo General11C.4.220644SecurityS-MPM2273Strutt,GuenaelNo Technical11C.3.220359SecurityS-MPM2274Strutt,GuenaelNo General11C.4.3.420949SecurityS-MPM2275Strutt,GuenaelNo General11C.4.3.320918SecurityS-MPM2276Strutt,GuenaelNo General11C.4.3.220844SecurityS-MPM2277Strutt,GuenaelNo Technical11C.32039SecurityS-MPM2278Sakoda,KazuyukiNo Technical11C.82311RFI R-LM2279Sakoda,KazuyukiNo Technical11.319152GeneralG-General2280Sakoda,KazuyukiNo Technical10.3.85.2.218325RFI R-LM2281Sakoda,KazuyukiNo Technical9.9.1.211232MACM-General2282Sakoda,KazuyukiNo Technical7.4.15.17513GeneralG-Frame2283Sakoda,KazuyukiNo Technical7.3.2.1166534RFIR-Proxy2284Sakoda,KazuyukiNo Technical7.2.3.103259SecurityS-General2285Sakoda,KazuyukiNo Technical5.2.3.1721GeneralG-Base2286Sakoda,KazuyukiNo Editorial 5.2.14.4108GeneralG-Base2287Sakoda,KazuyukiNo Editorial 5.2.14.498GeneralG-Base2288Malinen,JouniNo Editorial8.2a.7.2.49351SecurityS-General2289Malinen,JouniNo Technical7.3.23761MACM-MCCA2290Housley,RussellYes Technical11C.3.2.1SecurityS-General2291Mccann,StephenNo Technical11.23.61976GeneralG-EmergencyComment Proposed Change Resoluti on StatusNames of state machines do not need to be in caps.Replace the figure name with "Table 11C-3--Finite state machine of the authenticated mesh peering exchange protocol"Names of state machines do not need to be in caps.Replace the table name with "Table 11C-2--Authenticated mesh peering exchange finite state machine"English clean up.Replace "element" with "elements"after both "RSN" and "AMPE".Insert "the" before "Authenticated"and, since the following terms do notrefer to the field names, on lines 61through 64 replace "AuthenticatedMesh Peering Exchange" with"authenticated mesh peeringexchange".(twice) and "Mesh PeeringManagement" with "mesh peeringThe capitalized names do not refer to frame, field, primitve, etc. names.Replace "Authenticated Mesh Peering Exchange" with "authenticated mesh peering exchange".and "Mesh Peering Management" with "mesh peering management" throughout the draft when the terms do not directly refer to frames, fields, primitive names, etc.In this heading "exchange" is not Replace "Mesh Peering Management"with "mesh peering management". The indicator "the" for the peer STAbrings up the question "which one?".Replace "the peer" with "its peer".Need colon.Place a colon at the end of the line. Passive.Replace "When verifying .. followed:"with:"When the mesh STA verifies a MeshPeering Management frame, it shallfollow the procedure:"Need period.Place a period at the end of the line. Need colon.Place a colon at the end of the line.Passive.Replace "When constructing ...followed:" with "When the mesh STAconstructs a Mesh PeeringManagement frame, it shall follow theRequirements can't be stated in informative notes, and "required" and "must" are both deprecated in IEEE Replace "NOTE--" with "Note that". Replace "is required to" with "shall". Replace "must" with "shall".Need a pause after "FAIL".Insert a comma after "FAIL". Subortinate clause needs a comma.Insert a comma after "authenticating"In references to action frames, the word "Action" is not capitalized.Replace "Action" with "action" on both lines 29 and 40."protection in the Mesh Peering Open action frame' sounds like the protection is inside the frame.Replace this complete paragraph with "A mesh STA shall announce in a Mesh Peering Open action frame the group cipher suite it uses for broadcast protection. When it receives a Mesh Peering Open frame from a candidate peer, it shall verify that it supports the candidate's announced group cipher suite . In addition, if the mesh STA receives a Mesh Peering Confirm frame, it shall verify that it supports the group cipher suite listed in that frame. If either selection fails, the mesh STA shall issue the appropriate reply frame with the MESH-INVALID-SECURITY-CAPABILITY reason code.""selector of the selected"""if generating" is not clear enough.Insert "it is" before "generating". Passive.Insert "the mesh STA shall generatethe" before "failure" and delete "shallbe generated".Confusing writing Replace "the chosen pairwise ciphersuites as the result of step b). If theydo not match," with "the pariwisecipher suite chosen in step b). Ifthere is no match,".Passive.Insert "the mesh STA shall generatethe" before "failure" and delete "shallbe generated"."chosen" needs to be closer to the name of the entity doing the choosing.Move "chosen" from its current location to one immediately following "suite".Too many repetitions of "the mesh STA" for clarity.Replace "the mesh STA supports" with "it supports" (since it is clear that this reference is the decisionmaker mesh STA)."STA with the largest" is vague.Replace "with" with "that has" andreplace "in the lexicographic ordering"with "(in lexicographic order)"."not empty and contains more than oneentry" is redundant.Delete "is not empty and".Missing indicator and passive.Replace "and" with ", the mesh STAgenerates the" before "failure" andreplace "shall be generated andcorresponding actions shall be takenaccording to" with " and then takesthe corresponding actions specified English clean up.Replace "shall independently makedecision on" with "shall make itsdecision about" and "based onintersection" with "based on theintersection".Missing indicator.Insert "the" before "last".Bulky writing.Replace "most preferred cipher suiteby the mesh STA" with "mesh STA'smost preferred cipher suite".In references to action frames, the word "Action" is not capitalized.Replace "Action" with "action" on both lines 29 and 40.Misplaced apostrophe.Replace "STAs'" with "STA's" as thisrefers to a single STA. Alternatively,could replace just with "STA"."Mesh" in "Mesh TKSA" does not need to be capitalized.Replace "Mesh TKSA" with "mesh TKSA"."via the active authenticaion protocol" is confusing at best.Delete "via the active authentication protocol" as that concept is incorporated in "initiation of the protocol".In this heading "Authenticated Mesh Peering Management" does not refer to a frame, field, primitive name, parameter name, etc.Replace "Authenticated Mesh Peering Exchange" with "Authenticated mesh peering exchange" throughout the draft, whenever this term does not apply directly to a frame, field, primitive name, etc."NOTE--" is used only with informative statements, while "is required to" and "must" indicate normative behavior.Replace "NOTE--" with "Note that". Replace "is required to" with "shall". Replace "must" with "shall".Antecedent clause needs a comma.Insert a comma after "STA". "must" is deprecated in IEEE standards.Replace "must be" with "is". Antecedent clause needs a comma.Insert a comma after "AMPE". "must" is deprecated in IEEE standards.Replace "must" with "shall". "must" is deprecated in IEEE standards.Replace "must" with "shall".In the 802.11 naming style "MLME-" primitive names are in all-caps.Replace "MLME-MWMPMeshPathSelection" with "MLME-MWMPMESHPATHSELECTION"In the 802.11 naming style "MLME-" primitive names are in all-caps.Replace "MLME-MeshLinkMetricReport" with "MLME-MESHLINKMETRICREPORT" throughout the draft.In the 802.11 naming style "MLME-" primitive names are in all-caps.Replace "MLME-MeshLinkMetricRead" with "MLME-MESHLINKMETRICREAD" throughoutIn the 802.11 naming style "MLME-" primitive names are in all-caps.Replace "MLME-MBSSGateAnnouncement" with "MLME-MBSSGATEANNOUNCEMENT"In the 802.11 naming style "MLME-" primitive names are in all-caps.Replace "MLME-MBSSProxyUpdate" with "MLME-MBSSPROXYUPDATE" throughout the draft.In the 802.11 naming style "MLME-" primitive names are in all-caps.Replace "MLME-MBSSCongestonControl" with "MLME-MBSSCONGESTIONCONTROL"。

无尺度网络摘自《科学美国人》中文版2003.7网络有随机网络和无尺度网络，许多网络包括因特网"人类社会和人体细胞代谢网络等，都是无尺度网络。

研究无尺度网络，对于防备黑客攻击、防治流行病和开发新药等，都具有重要的意义。

(原文:Scale-F ree Netw orks, pp50-59, May2003) 撰文/A lbert-Laszlo Barabasi, Eeic Bonabeau作者介绍A lbert-Laszlo Barabasi和Eric Bonabeau研究了从因特网到昆虫群落等一系列复杂系统的行为和特性。

Barabasi是美国圣母大学的霍夫曼物理学教授。

并在校内指导对复杂网络的研究，他著有《连结:网络新科学》一书。

Bonabeau为美国麻省剑桥咨询公司"伊可系统"的首席科学家，专门运用复杂科学方面的工具来开发商业机会。

他与别人合作撰写了《虫群智慧:从自然系统到人工系统》一书。

这是他在本刊上第二次发表文章。

一个实例：如图所示，因特网是一个无尺度网络，其中某些站点似乎与无数的其他站点相连结(参见右图的星爆形结构细节)。

本图绘制于2003年2月6日，描绘了从某一测试站点到其他约10万个站点的最短连结路径。

图中以相同的颜色来表示相类似的站点。

大脑，是由轴突相连结的神经细胞网络，而细胞本身，又是由生化反应相连结的分子网络。

社会也是一个网络，它由友情、家庭和职业关系彼此连结。

在更大的尺度上，食物链和生态系统可以看作由物种所构成的网络。

科技领域的网络更是随处可见:因特网、电力网和运输系统都是实例。

就连在文章中我们用以向你传递思想的语言，也是一种藉由语法相互串连在一起的文字网络。

尽管网络是如此重要和普遍，但科学家对它的结构和属性却知之不多。

在复杂的基因网络中，故障节点是如何相互作用而引发癌症的?在特定的社会和通信系统中，疾病和电脑病毒如何快速传播而导致流行?某些网络即便大部分节点失效，还能维持运行，原因何在?最近的研究开始找到这些问题的答案。

参考文献标准格式来源：赵现勇的日志参考文献类型：专著［m］，论文集［c］，报纸文章［n］，期刊文章［j］，学位论文［d］，报告［r］，标准［s］，专利［p］，论文集中的析出文献［a］电子文献类型：数据库［db］，计算机［cp］，电子公告［eb］电子文献的载体类型：互联网［ol］，光盘［cd］，磁带［mt］，磁盘［dk］a：专著、论文集、学位论文、报告［序号］主要责任者.文献题名［文献类型标识］.出版地：出版者，出版年:起止页码（可选）［1］刘国钧，陈绍业.图书馆目录［m］.北京：高等教育出版社，1957.15-18. b:期刊文章［序号］主要责任者.文献题名［j］.刊名，年，卷（期）：起止页码［1］何龄修.读南明史［j］.中国史研究，1998,(3):167-173.［2］ou j p，soong t t，et al.recent advance in research on applications of passiveenergy dissipation systems［j］.earthquack eng,1997,38(3):358-361. c:论文集中的析出文献［序号］析出文献主要责任者.析出文献题名［a］.原文献主要责任者（可选）.原文献题名［c］.出版地：出版者，出版年.起止页码［7］钟文发.非线性规划在可燃毒物配置中的应用［a］.赵炜.运筹学的理论与应用——中国运筹学会第五届大会论文集［c］.西安：西安电子科技大学出版社，1996.468. d：报纸文章［序号］主要责任者.文献题名［n］.报纸名，出版日期（版次）［8］谢希德.创造学习的新思路［n］.人民日报，1998-12-25（10）. e：电子文献[文献类型/载体类型标识]：[j/ol]网上期刊、[eb/ol]网上电子公告、[m/cd]光盘图书、[db/ol]网上数据库、[db/mt]磁带数据库［序号］主要责任者.电子文献题名［电子文献及载体类型标识］.电子文献的出版或获得地址，发表更新日期/引用日期［8］万锦.中国大学学报文摘（1983-1993）.英文版［db/cd］.北京：中国大百科全书出版社，1996篇二：参考文献书写格式参考文献书写格式文后参考文献著录格式(电子版) a.连续出版物［序号］主要责任者．文献题名［j］．刊名，出版年份，卷号(期号)：起止页码．［１］袁庆龙，候文义．ni-p合金镀层组织形貌及显微硬度研究［ｊ］．太原理工大学学报，2001，32(1)：51-53.b.专著［序号］主要责任者．文献题名［ｍ］．出版地：出版者，出版年：页码．［３］刘国钧，郑如斯．中国书的故事［ｍ］．北京：中国青年出版社，1979：115．c.会议论文集［序号］析出责任者．析出题名[a]．见(英文用in)：主编．论文集名[c]．(供选择项：会议名，会址，开会年)出版地：出版者，出版年：起止页码．［６］孙品一．高校学报编辑工作现代化特征［a］．见：中国高等学校自然科学学报研究会．科技编辑学论文集(2)[c]．北京：北京师范大学出版社，1998：10-22．d.专著中析出的文献［序号］析出责任者．析出题名[a]．见(英文用in)：专著责任者．书名[m]．出版地：出版者，出版年：起止页码．［12］罗云．安全科学理论体系的发展及趋势探讨[a]．见：白春华，何学秋，吴宗之．21世纪安全科学与技术的发展趋势[m]．北京：科学出版社，2000：1-5．e.学位论文［序号］主要责任者．文献题名［ｄ］．保存地：保存单位，年份：［７］张和生．地质力学系统理论［ｄ］．太原：太原理工大学，1998：f.报告［序号］主要责任者．文献题名［ｒ］．报告地：报告会主办单位，年份：［９］冯西桥．核反应堆压力容器的lbb分析［ｒ］．北京：清华大学核能技术设计研究院，1997： g.专利文献［序号］专利所有者．专利题名［p］．专利国别：专利号，发布日期：［１１］姜锡洲．一种温热外敷药制备方案［ｐ］．中国专利：881056078，1983-08-12：h.国际、国家标准［序号］标准代号．标准名称［s］．出版地：出版者，出版年：［1］gb/t 16159—1996．汉语拼音正词法基本规则［s］．北京：中国标准出版社，1996：i.报纸文章［序号］主要责任者．文献题名［ｎ］．报纸名，出版年，月(日)：版次．［１3］谢希德．创造学习的思路[ｎ]．人民日报，1998，12(25)：10 j.电子文献［序号］主要责任者．电子文献题名［文献类型/载体类型］．：电子文献的出版或可获得地址(电子文献地址用文字表述)，发表或更新日期/引用日期(任选) ：［21］姚伯元．毕业设计(论文)规范化管理与培养学生综合素质［eb/ol］．：中国高等教育网教学研究，2005-2-2：附：参考文献著录中的文献类别代码普通图书：m 会议录：c 汇编：g 报纸：n 期刊：j 学位论文：d 报告：r标准：s 专利：p 数据库：db 计算机程序：cp 电子公告：eb 中华人民共和国国家标准gb7714—87《文后参考文献著录规则》中规定：“引用的文献的标注方法可以采用顺序编码制，也可以采用‘著者—出版年制”。

Bio Med CentralBMC Family PracticeStudy protocolRandomized double-blind placebo-controlled trial of sublingualimmunotherapy in children with house dust mite allergy in primary care: study design and recruitmentCindy MA de Bot*1, Heleen Moed 1, Marjolein Y Berger 1, Esther Röder 1,2, Hans de Groot 2, Johan C de Jongste 3, Roy Gerth van Wijk 2 and Johannes C van der Wouden 1Address: 1Department of General Practice, Erasmus MC-University Medical Center Rotterdam, PO Box 2040, 3000 CA Rotterdam, the Netherlands, 2Department of Allergology, Erasmus MC-University Medical Center Rotterdam, PO Box 2040, 3000 CA Rotterdam, the Netherlands and3Department of Pediatric Respiratory Medicine, Erasmus MC-University Medical Center/Sophia Children's Hospital, Rotterdam, PO Box 2040, 3000 CA Rotterdam, the NetherlandsEmail: Cindy MA de Bot*-c.debot@erasmusmc.nl; Heleen Moed -h.moed@erasmusmc.nl; Marjolein Y Berger -m.berger@erasmusmc.nl; Esther Röder -e.roder@erasmusmc.nl; Hans de Groot -grooth@rdgg.nl; Johan C de Jongste -j.c.dejongste@erasmusmc.nl; Roy Gerth van Wijk -r.gerthvanwijk@erasmusmc.nl; Johannes C van der Wouden -j.vanderwouden@erasmusmc.nl * Corresponding authorAbstractBackground: For respiratory allergic disorders in children, sublingual immunotherapy has been developed as an alternative to subcutaneous immunotherapy. Sublingual immunotherapy is more convenient, has a good safety profile and might be an attractive option for use in primary care. A randomized double-blind placebo-controlled study was designed to establish the efficacy of sublingual immunotherapy with house dust mite allergen compared to placebo treatment in 6to18-year-old children with allergic rhinitis and a proven house dust mite allergy in primary care. Described here are the methodology, recruitment phases, and main characteristics of the recruited children.Methods: Recruitment took place in September to December of 2005 and 2006. General practitioners (in south-west Netherlands) selected children who had ever been diagnosed with allergic rhinitis. Children and parents could respond to a postal invitation. Children who responded positively were screened by telephone using a nasal symptom score. After this screening, an inclusion visit took place during which a blood sample was taken for the RAST test.Results: A total of 226 general practitioners invited almost 6000 children: of these, 51% was male and 40% <12 years of age. The target sample size was 256 children; 251 patients were finally included. The most frequent reasons given for not participating were: absence or mildness of symptoms, absence of house dust mite allergy, and being allergic to grass pollen or tree pollen only. Asthma symptoms were reported by 37% of the children. Of the enrolled children, 71% was sensitized to both house dust mite and grass pollen. Roughly similar proportions of children were diagnosed as being sensitized to one, two, three or four common inhalant allergens.Conclusion: Our study was designed in accordance with recent recommendations for research on establishing the efficacy of sublingual immunotherapy; 98% of the target sample size was achieved. This study is expected to provide useful information on sublingual immunotherapy with house dust mite allergen in primary care. The results on efficacy and safety are expected to be available by 2010.Trial registration: the trial is registered as ISRCTN91141483 (Dutch Trial Register)Published: 20 October 2008BMC Family Practice 2008, 9:59doi:10.1186/1471-2296-9-59Received: 21 April 2008Accepted: 20 October 2008This article is available from: /1471-2296/9/59© 2008 de Bot et al; licensee BioMed Central Ltd.This is an Open Access article distributed under the terms of the Creative Commons Attribution License (/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.BackgroundSpecific immunotherapy with allergens might prevent the onset of asthma in individuals with allergic rhinitis and may accelerate the remission of asthma in children with allergic disease. [1-3] Although subcutaneous immuno-therapy (SCIT) is an effective treatment of respiratory allergic disorders, [4] the injections can be uncomfortable and side effects, though rare, may be serious and even fatal. [5,6] The use of specific sublingual immunotherapy (SLIT) for treatment of respiratory allergic disorders in children may be a viable alternative to SCIT because of its convenient form of administration and good safety pro-file – which has allowed home administration of SLIT. [7,8] Thus, although SLIT seems particularly suitable for children in primary care, most clinical trials up to now have been performed in a hospital setting.Evidence for the efficacy of SLIT in children remains inconclusive. Various reviews concluded that there was insufficient evidence to recommend SLIT for use in rou-tine clinical practice. [9-11] In their Cochrane review, Wil-son et al. concluded that SLIT is an accepted treatment for adults; studies with children revealed no significant reduc-tion in symptoms and medication scores, but the number of participants was small. [12]In 2001, the Allergic Rhinitis and its Impact on Asthma (ARIA) guidelines were published in co-operation with the World Health Organization. [13] They recommend treatment of allergic rhinitis in a stepwise manner (using a combination of allergen avoidance, pharmacotherapy and immunotherapy) based on the duration and severity of disease, rather than on the type of exposure (i.e. sea-sonal, perennial, occupational) as in previous guidelines.[14] Immunotherapy is recommended for patients with more severe disease, for those not responding to usual treatments, or for those refusing usual treatments; this type of patient is generally treated in a hospital setting and/or by a specialist.In the Netherlands, allergic rhinitis in children is usually managed by the general practitioner (GP). We hypothe-sized that SLIT could be an effective treatment in primary care and designed a study to evaluate the efficacy and safety of SLIT in children and adolescents with house dust mite-induced allergic rhinitis. Here we describe the meth-odology, recruitment, and main characteristics of the pri-mary care study population.MethodsStudy designThis ongoing study is a randomized double-blind pla-cebo-controlled study, comparing the efficacy of SLIT with house dust mite allergen (SLIT-HDM) to that of placebo treatment in 6 to 18-year-old children with allergic rhini-tis and a proven house dust mite allergy in primary care. Patients entered the study and started treatment either in September-December 2005 or in September-December 2006 for a period of approximately two years. Written informed consent was obtained. The study was approved by the Ethical Review Board of Erasmus MC-University Medical Center Rotterdam. The trial was registered as ISRCTN91141483.Participants and recruitmentGPs in south-western Netherlands selected children aged 6 to 18 years in their computerized patient files with either a diagnosis of hay fever/allergic rhinitis or relevant medication use: i.e. antihistamines for systemic use; nasal corticosteroids; topical decongestants; topical anti-aller-gics, and other nasal preparations.Recruitment took place September to December in 2005 and in 2006. An information letter signed by the GP was sent to the selected children. This letter described the gen-eral purpose of the study, elicited cooperation, and pro-vided a return form and envelope. On the return form children and parents could indicate whether or not they were interested in the study; if not interested they could indicate the reason for not participating.Participants who responded positively were telephoned by a research assistant to arrange a screening interview (see below). The research assistant asked questions about nasal symptoms during the last three months, the history of allergic rhinitis, general medication use, and use of asthma medication. Table 1 gives an overview of all inclu-sion and exclusion criteria.After telephone screening an inclusion visit took place for those who met the inclusion/exclusion criteria and who agreed (children/parents) to further participation. During this visit, the research assistant performed/recorded the following: rhinitis symptoms during the last month and last week (nasal symptoms: rhinorrhea, blocked nose, sneezing, itching); conjunctivitis symptoms during the last month and last week (eye symptoms: tearing, itching, redness); International Study of Asthma and Allergies in Childhood (ISAAC) questionnaire [15] for rhinitis and asthma; wheeze and cough; family history of allergy, asthma and eczema; rhinoconjunctivitis-specific quality of life for pediatrics and adolescents (PRQLQ and AdolR-QLQ[16,17]); blood sample for RAST (grass pollen, tree pollen, HDM, cat dander and a pet, if present at home) (CAP-Phadiatop®, Pharmacia Diagnostics AB, Uppsala, Sweden); and physical examination (weight and height). After the screening visit, when children met the inclusion criteria and none of the exclusion criteria and children and parents agreed to participate, a home visit was sched-uled to provide instructions about the baseline diary. Every day for one month, children recorded the symp-toms related to allergic rhinitis on a diary card; also reported were other complaints, rescue medication, and other medication needed (see below). At this visit the research assistant took dust samples from the child's bed-room floor and mattress to assess indoor HDM exposure. This will be repeated after two years.After the baseline diaries had been completed a new visit was scheduled and, after signing informed consent, partic-ipants were assigned to SLIT treatment or placebo accord-ing to the randomization schedule (see below). RandomizationRandomization was generated by a computer program in varying block sizes unknown to the investigators. The ran-domization list was passed to the Department of Phar-macy at Erasmus MC. In order to ensure that disease severity was similar between patients assigned to verum therapy and those assigned to placebo, randomization was stratified according to severity on the basis of data obtained during the telephone screening.InterventionParticipants received an aqueous extract of house dust mites (Dermatophagoides pteronyssinus) in a glycerinated isotonic phosphate buffered solution (Oralgen Mijten, Artu Biologicals, Lelystad, the Netherlands) or placebo treatment consisting of the glycerol solvent. In accordance with the manufacturer's guidelines the treatment period was divided into two phases: a dose escalation phase of 20 days, and a maintenance phase of approximately two years. Treatment started on day one with a single drop. One drop consisting of 0.05 ml corresponds with 35 bio-logical units (BU); the dose was increased by one drop per day until day 20 (20 drops = 1 ml = 700 BU). The mainte-nance dose was 20 drops (= 700 BU) twice weekly. The drops were administered sublingually and kept there for at least 1 minute before being swallowed. A research assistant instructed the participants and also provided written instructions. Participants, parents, investigators, research assistants and caregivers were blinded to treat-ment allocation.Follow-upFigure 1 shows the time schedule per individual patient. After randomization children started with treatment for 20 days (dose escalation phase) followed by a mainte-nance phase of two years. Children filled in a diary during three months (between September and December) after one and two years of treatment (see below). Every month a research assistant completed a questionnaire (con-ducted by telephone) throughout the entire study period. Over the two years of treatment the total number of planned contacts is 13 home visits and 23 telephone calls. Outcome measuresThe primary outcome measure for efficacy is the differ-ence between the group receiving SLIT and the group receiving placebo for the total daily mean rhinitis symp-tom score for four nasal symptoms (see below), assessed through a diary filled in during three months after two years of treatment. In the period of evaluation (September through December), the percentage of days on which the daily symptoms are properly recorded should be at least 50%. For patients who do not meet this criterion in the second year (e.g. dropouts after 1 year) data of the first treatment year will be used. See section Data analysis for further details.Secondary outcome measures are the difference between the group receiving SLIT and the placebo group for the proportion of symptom-free days, the proportion of res-cue medication-free days, use of rescue medication, mean eye symptom score, total symptom score (nasal and eyeTable 1: Inclusion and exclusion criteria for the study populationInclusion criteria• aged 6–18 years• history of allergic rhinitis for at least 1 year• IgE antibodies ≥0.7 kU/l to house dust mite• no use of nasal steroids in the month before start of baseline measurements• rhinitis symptom score of at least 4 out of 12 during last 3 months• signed informed consentExclusion criteria• severe asthma(requiring 800 mcg budesonide daily or equivalent for other inhaled steroids; or requiring >3 courses of oral prednisone/prednisolone in previous year or required hospital stay for asthma in previous year)• sensitization to pets present at home (IgE antibodies ≥0.7 kU/l)• planned surgery of nasal cavity• having received immunotherapy in past 3 years• language barrier• contraindications to sublingual immunotherapy (as supplied by the manufacturer)Figure 1Detailed time schedule per individual patient.symptoms), and disease-specific quality of life after two years of treatment. Overall evaluation of the treatment effect will be assessed by patient, parents and research assistant after two years of treatment.Assessment of efficacyEfficacy will be measured by patient-assessed symptom scores. Although nasal, eye, skin and lung-related symp-toms have been related to house dust mite allergy, the main allergic symptoms are considered to be the follow-ing nasal symptoms: sneezing, itching, watery running nose and blockage. The intensity of these symptoms is subjectively assessed according to a grading scale: 0 = no complaints, 1 = minor complaints, 2 = moderate com-plaints and 3 = serious complaints; the maximum score is 12. The scores will be assessed daily by the patient and recorded in the patient's diary. The period of measure-ment will be three months in the period September through December in 2006 and 2007 for the primary out-come measures (first cohort), and in 2007 and 2008 (sec-ond cohort); this autumnal period of the year was chosen because it has the highest HDM exposure levels. Assessment of safety, tolerability and compliance Adverse effects will be assessed by patients and parents reporting effects in the diary, or calling the research assist-ant with complaints, or by the research assistant via a questionnaire filled in during home visits, and by monthly telephone contact. All adverse events reported during the study will be recorded. In case of serious adverse events or persisting allergic symptoms after man-agement according to protocol, the study treatment will be discontinued for these patients. If patients discontinue the study medication, they will be asked to agree to further follow-up according to the study protocol during the remainder of the study period.Compliance will be measured by self-report of SLIT administration in the diary and by monthly telephone contact, and determined by weighing the returned study medication.Sample size calculationAs rhinitis symptoms are the primary outcome measure, this was used for calculating the sample size. A Dutch study on mattress covers provides relevant data for symp-tom scores in patients with house dust mite allergy (aged 8–50 years). [18] Based on the baseline symptom score in the latter study, and the ability to assess a reduction of at least 30% (proposed by Malling as a clinically relevant reduction) [19], in our study a sample size of 96 patients per group would be required. Taking into account a drop-out rate of 25% between randomisation and end of fol-low-up, this would require 128 patients in each study group. An alternative approach is assuming the nasal score at the last week screening visit to be 4.5 (sd 2.6). A 30% change would provide a delta of approximately 0.5 (generally assumed to be clinically relevant) and require a sample size of 105 per study group (alpha = 0.05 and beta = 95%).Quality of lifeRhinoconjunctivitis-specific quality of life will be assessed through the validated Pediatric (6–11 years) and Adoles-cent (12–17 years) Rhinoconjunctivitis Quality of Life Questionnaire (PRQLQ and AdolRQLQ, respectively) at baseline and after one and two years. [16,17] To establish the presence of lower airway symptoms during the last 12 months at baseline, specific questions on wheezing and dry cough at night were taken from the ISAAC. [15] Rescue medicationDuring the study the use of symptomatic allergy medica-tion is discouraged, especially use of long-acting antihista-mines and locally or generally administered corticosteroids. However, rescue treatment is allowed in case of persisting allergic symptoms (levocetirizine tab-lets, xylomethazoline nasal spray and levocabastine eye drops); the above-mentioned rescue medication will be provided free of charge. In principle, patients are encour-aged to use the provided medication only, but are allowed to use their own medication as well. Patients were clearly instructed on the use of rescue medication and other med-ication, and on how to document entries in the patient diary. For severe or steadily worsening rhinoconjunctivitis symptoms or intermittent asthma the patient should con-sult his/her physician.Data analysisThe treatment effect will be tested at a two-sided signifi-cance level of 5%. Statistical comparison between verum and placebo of the mean daily sum score from the diary after two years will be done using Analysis of Covariance. There will be three covariates in this analysis: baseline nasal sum score at entry into study, age of patients, and presence of cat allergy. In case more than one child from the same family has been included in the study and con-tribute to the final analysis, we will test whether 'family' provides a statistically significant effect (P < 0.20). In that case 'family' will be added as a random effect. Exploratory subgroup analyses are planned for the differ-ence between placebo and verum regarding the primary outcome according to age and the baseline symptom score (both dichotomized at the median value).All analyses will be performed according to the intention-to-treat principle, i.e. irrespective of compliance with the prescribed dosing schedule and other treatments, but excluding patients in whom major inclusion criteria werenot fulfilled. A per-protocol analysis will include all patients who took at least 80% of the study medication and completed 50% of the diaries.For this paper, the distribution of age and gender through-out the recruitment period will be compared. All data are presented as summary descriptive statistics: means, stand-ard deviations (SD) or percentages. Statistical analyses were carried out with SPSS version 11.0 and differences of p < 0.05 were considered significant.ResultsLetters were posted by 226 general practitioners to 5986 children. An answer form was returned by 2555 children; of these, 1072 children responded positively to the letter and 500 of these children were included after the screen-ing by telephone. Finally, 251 children (i.e. only 4.2% of the children selected in general practice) were included in the study.Table 2 summarizes the main reasons given for not partic-ipating in the consecutive recruitment phases. In response to the initial mailing most of those who declined had few or no complaints (48%), or had another allergy (16%). During the telephone screening, those not included had no history of HDM allergy (28%) or a low symptom score (28%). In the last phase of the recruitment (the screening visit) the main reasons for non-participation were no HDM allergy but only grass or tree pollen allergy detected by RAST (33%), and no sensitization to inhalant allergens detectable by RAST (30%).Table 3 presents the baseline characteristics of the included patients. The mean age of the participants was 11.8 (SD 3.0) years. A total of 251 children were rand-omized to treatment or placebo. During the recruitment period nasal complaints were assessed at several time points; this symptom score showed a difference between telephone screening (6.8) and screening visit (4.5). More than half of the children reported wheeze/breathlessness (54%) and dry cough (53%) during the last year. In almost 37% of the children asthma was reported.The majority of the children (77%) were multisensitized. Roughly similar proportions of children were diagnosed as being sensitized to one, two, three or four common allergens. Of the included children, 71% was sensitized to both HDM and grass pollen, followed by tree pollen in 43%, and cat dander in 34% of the children.Table 4 shows the distribution of age and gender during the recruitment process. Of almost 6,000 children, 51%Table 2: Reasons not to participate in the consecutive recruitment phasesReasons not to participate Total(n)PercentageLetter returned (n = 1483)Few or no complaints71048.4% Other allergy24016.4% Study too burdensome20213.8% No interest in the study18612.6% No reason1459.8% Telephone screening (n = 572)No HDM allergy15927.8% Low symptom score (<4/12)15827.6% Not interested in study5710.0% Severe asthma39 6.8% Language barrier27 4.7% Use of immunotherapy in the last 3 years19 3.3% Refusing blood sample to be taken17 3.0% Age (out of range)15 2.6% Allergic complaints <1 year12 2.1% History of severe allergic reaction9 1.6% Systemic disease8 1.4% Use of nasal corticosteroids 1 month before baseline7 1.2% Answer forms received after deadline of inclusion period457.9% Screening visit (n = 249)Only grass pollen or tree pollen sensitization8132.5% No sensitization detectable7530.1% Sensitive to pet at home (confirmed by RAST)6024.1% No informed consent2911.7% Use of unallowed co-medication4 1.6%was male and 40% was aged 6–11 years. In the final recruitment phase, 251 children were included in the study. The distribution of age (6–11 years, p = 0.006) and gender (boys 59%, p < 0.025) of the children included in the present study is significantly different from those who initially received the invitation letter.DiscussionThis is an ongoing randomized double-blind placebo-controlled trial to establish the efficacy of sublingual immunotherapy with house dust mite allergen in children in primary care. Because the effectiveness of SLIT is still under discussion (mainly due to inconclusive quality/ methodology of the published trials), the present long-term study is expected to provide useful information about SLIT with house dust mite allergen in primary care. Although the distribution of age and gender of the partic-ipating children is significantly different from those con-tacted in the first recruitment phase, the difference is relatively small and age and gender groups are adequately represented; therefore, this difference should not affect the generalizability of the results of the trial.Table 3: Baseline characteristics of the included childrenTotal(n = 251)PercentageGenderMale14959.4% Female10240.6% AgeMean (SD) in years: 11.8 (3.0)6–11 years12248.6% 12–17 years12951.4% Physical characteristicsWeight in kg: mean (SD)47.5 (15.3)Height in cm: mean (SD)154.6 (17.1)Season with most complaints of allergySpring3513.9% Autumn14 5.6% Spring and autumn/entire year20180.1%Nasal symptoms (scale 0–12)Telephone screening: mean (SD) 6.8 (2.1)Screening visit in last 3 months: mean (SD) 5.8 (2.3)Screening visit in last week: mean (SD) 4.5 (2.6)AsthmaAsthma present9236.7% Asthma medication9939.4%Wheeze/breathless – ever15462.3% Wheeze/breathless – last year13153.9% Dry cough at night – last year13052.6%SensitizationOne allergen (monosensitized for HDM)5823.1% Two allergens6726.7% Three allergens7228.7% Four allergens5421.5%Sensitization to both HDM andGrass pollen17971.3% Tree pollen10843.0% Cat dander8533.9%Strengths and weaknessesThe importance of the methodology and quality of immu-notherapy trials has been documented. [19] The present study has a baseline assessment and complies with other recommendations: i.e. placebo-controlled, double-blind, randomized, adequate sample, sufficient duration of treatment, patients selected according to predefined clini-cal criteria, and clearly defined primary and secondary outcomes.Most related studies have been performed in a hospital setting, [20,21] so that the results may not be applicable to the general population. Therefore, our study is designed to evaluate – in a primary care setting – the effi-cacy and safety of SLIT in children and adolescents with house dust mite-induced allergic rhinitis.The ARIA guidelines propose that SLIT can be adminis-tered to young patients if these children are carefully selected with rhinitis, conjunctivitis and/or asthma caused by pollen and mite allergy. [13] By recruiting young chil-dren from a primary care setting (according to our meth-odology) the included children will meet this recommendation.Most earlier studies failed to report on the phase prior to randomization, whereas the present study reports the rea-sons given not to participate and possible selection bias. According to the WAO Task Force, the ideal efficacy study of specific allergen immunotherapy should be performed in monosensitized patients or in patients concomitantly sensitized to noncross-reacting allergens. [22] It is reported that single-allergen-specific immunotherapy may prevent sensitization to other airborne allergens in monosensitized children. [1,3,23] In our study we included both monosensitized and multisentized chil-dren; the majority was multisensitized and only 23% was monosensitized. We believe that this will increase the gen-eralizability of the study results to a wider range of patients.Many clinical trials face recruitment problems and have to approach many patients in order to include only a small proportion. [24,25] In a survey of 78 studies in Dutch pri-mary care, a median of 87% of planned patients was recruited. [26] In the present study 98% of the target sam-ple size was recruited.ConclusionOur study was designed in accordance with recent recom-mendations for research on establishing the efficacy of sublingual immunotherapy; 98% of the target sample size was reached. This study is expected to provide useful information on the position of SLIT with house dust mite allergen in primary care; results on the efficacy and safety of SLIT should be available by 2010.Competing interestsAs employees of Erasmus MC, MYB and JCvdW obtained research funding from Artu Biologicals, which included the present study. CMAdB, HM and ER were employed by Erasmus MC through these funds. JCdJ received project funding from GlaxoSmithKline, Merck, Sharpe & Dohme, Roche and Friso Foods, and lectured at scientific meetings on request of G laxoSmithKline, Novartis and Merck, Sharpe & Dohme. All payments went to the Erasmus Uni-versity Medical Center, Pediatric Research Holding. H. de G root declares to have no conflict of interest with the pharmaceutical world. R. Gerth van Wijk received fees for lectures and expert panel participation from Allmiral, Alcon, Merck Sharp & Dome, Novartis, Stallargènes and UCB.Authors' contributionsCMAdB carried out all analyses and drafted the manu-script. All other authors read and approved the final pro-tocol and the manuscript. JCvdW drafted the protocol and supervised the study.AcknowledgementsThe study was financially supported by Artu Biologicals, Lelystad, The Netherlands.Table 4: Distribution of age and gender during the recruitment phasesTotal Male Age group6–11 yearsn n%n% Total mailed5986306651.2%236939.6% Letter returned (irrespective of answer)2555133152.1%103640.5% Letter returned positive response107259255.2%47143.9% Telephone screening positive50027955.8%21442.8% Screening visit positive25114959.4% 112248.6%2 1: p = 0.006 (compared with 5986 children who were initially contacted)2: p = 0.025 (compared with 5986 children who were initially contacted)。

Unit 2Bards of the InternetTeaching PlanLearning Objectives1)Rhetorical skill: evidence and quotation2)Key language & grammar points3)Writing strategies: quotations to support opinions4)Theme: literacy and the InternetPre-class Activity: none.Relationship to Current Unit:Materials: Teacher’s Book (6), English-English Dictionary, Blackboard, PPTEstimated Time of Lesson: 270m, 6 sessionsTime Allocation:P 1-21.Pre-reading: playing games / word puzzle / picture activation / short stories 10m2.Global Reading: text introduction, culture notes, author, structure 15m3.Detailed Reading (a): Text I: Paragraph 1-3 65mP 3-44.Detailed Reading (b): Text I: Paragraph 4-13 90mP 5-65.Consolidation Activities (a): Text Comprehension; Writing Strategies 20m6.Consolidation Activities (b): Language work; Oral Activities; Writing 70m7.Further Enhancement (Optional): Text II / Other Comprehensive PracticesSection One Pre-reading ActivitiesI. Picture ActivationWhat does this picture imply?II. Pre- reading Questions1. We are entering an era of information explosion which is boosted by advanced science and technology. We have become accustomed to communicate with each other through Twitter, Weibo, Facebook, etc. The Internet has changed our life one way or another. Please share an example on how the Internet has changed your life or some habits.Open to discussion.2. The Internet has provided us with a more speedy and convenient means of communication, which is called for by the quickened pace of life today. Online communication has bred a style of writing with numerous distinctive features of its own. Can you think of some examples to illustrate the differences between online writing and traditional writing?Open to discussion.Section Two Global ReadingI. Text IntroductionIn this essay the author describes the current fashion for netwriting, explores the causes of the poor quality of writing found on the Internet, and explains its merits and differences from published writing as well as justifications for its survival and prevalence.II. Culture NotesCompuServe (Paragraph 2) Founded in 1969 as a computer time sharing service, CompuServe drove the initial emergence of the online service industry. In 1979, CompuServe became the first service to offer electronic mail capabilities and technical support to personal computer users. CompuServe broke new ground again in 1980 as the first online service to offer real-time chat. Since its acquisition by AOL in 1998, CompuServe has been a wholly owned subsidiary of America Online, Inc.Prodigy (Paragraph 2) an official web site, providing Internet access and related value-added servicesAmerica Online (Paragraph 2) the world’s largest online information service whic h is often abbreviated as AOL. It is an American corporation headquartered in Virginia, owner and operator of online and interactive computer games.McLuhan (Paragraph 2) Herbert Marshall McLuhan (1911–1980), Canadian writer and philosopher. He argued that it is the characteristics of a particular medium rather than the information it disseminates that influence and control society.Tor Books (Paragraph 3) Tor Books, an imprint of Tom Doherty Associates, LLC, is a New York-based publisher of hardcover and softcover books, founded in 1980 and committed (although not limited) to science fiction and fantasy literature. The company is now part of the Holtzbrink group, now known as Macmillan Publishers.Mark Twain (Paragraph 3)US novelist and humorist; pseudonym of Samuel Langhorne Clemens (1835–1910). After gaining a reputation as a humorist with his early works, he wrote his best-known novels, The Adventures of Tom Sawyer (1876) and The Adventures of Huckleberry Finn(1885); both give a vivid evocation of Mississippi frontier life.Tom Paine (Paragraph 3) Thomas Paine (1737–1809), 18th century English political writer. His pamphlet Common Sense (1776) called for American independence and The Rights of Man (1791) defended the French Revolution.Revolutionary War (Paragraph 3) indicating the war of American Revolution from 1775 to 1783 in which the American colonists won independence from British rule. The Revolutionary pamphleteers were not professional writers but common citizens engaged in the debate of ideas.Elizabethan era (Paragraph 3) Elizabeth (1533–1603), queen of England and Ireland (1558–1603); daughter of Henry VIII. She re-established Protestantism as the state religion. The phrase indicates the period of time reigned by Elizabeth.Gutenberg (Paragraph 3) Johannes Gutenberg (c. 1400–1468), 15th century German printer. He was the first in the West to print by using movable type and to use a press.Canterbury Tales (Paragraph 7) A collection of stories in poetic form written by Geoffrey Chaucer and published by William Caxton in 1478. Today, a visit to The Canterbury Tales, one of Kent’s most popular attractions, with its stunning reconstruction of 14th century England, is just like stepping into the Middle Ages. Here it refers to a very old-fashioned dressing style.WELL (Paragraph 9) a pioneering online community known for engaging conversations and intelligent debatesSausalito (Paragraph 9) a small city in northwestern California, across the Golden Gate from San Francisco, a noted artists colony with a population of about 8,000New York Times (Paragraph 9)one of the most important papers in the history of American newspapers, which was founded on September 18, 1851 by Henry Jarvis Raymond and George Jones. Raymond was also a founding director of the Associated Press in 1856.Wall Street Journal (Paragraph 9)Founded in 1889, the Wall Street Journal, the flagship publication of Dow Jones & Company, is the world’s leading business newspaper.Communications Daily (Paragraph 10)America’s pr eeminent daily telecom news source published by Warren Communications News, the publisher of must-read daily news publications covering the telecom and television industryDes Moines (Paragraph 12) the capital of Iowa, in the south central part of the state, with a population of about 200,000Brook Farm (Paragraph 13)an experimental commune in West Roxbury, Massachusetts, set up by a group of US writers in the 1840sthe Globe Theatre (Paragraph 13) a theatre in Southwark, London, erected in 1599, wher e many of Shakespeare’s plays were first publicly performedIII. AuthorPhillip Elmer-DeWitt is a senior editor of Time magazine. Since 1982, Elmer-DeWitt has written about science and technology for the magazine. As a staff writer for that publication, he has launched two new sections in the magazine: in 1982, “Computers” and in 1987, “Technology.” In 1993, Elmer-DeWitt also spearheaded Time Online, the interactive edition of the magazine. Elmer-DeWitt has produced over four hundred news and feature stories on subjects ranging from in vitro fertilization to computer sex. His Time magazine cover stories include “Computer Viruses” (1988), “Supercomputers” (1988), “Curing Infertility” (1991), “Cyberpunk” (1993), “Info Highway” (1993), “Video Games” (1993), “Human Cloning” (1994), “The Internet” (1994), and “Sex in America” (1994).IV. Structural AnalysisPart 1(paras 1-2) introductory leadPart 2(paras 3-4) feature and quality of net writingPart 3(paras 5-9) reasons and defensePart 4(paras 10-11) merits of net writingPart 5(paras 12-13) counter-argumentSection Three Detailed ReadingBARDS OF THE INTERNETPhillip Elmer-DewittI. AnalysisParagraphs 1-2 AnalysisThe first two paragraphs are an introductory lead-in that presents an analytical comparison between what happened to writing when the telephone was invented and what is happening on computer networks now.Paragraph 3 AnalysisIn this paragraph a number of people’s remarks are quoted. These quotations serve as evidence of the fact that deviations from the traditional medium of letter writing are not rare in history. The implication is that we should not be surprised by the boom in netwriting.Paragraph 4 AnalysisIn this paragraph the author concedes the most serious problem of netwriting — its poor quality.Paragraphs 5-6 AnalysisIn these two paragraphs the author explores the reasons why the quality of netwriting is so poor. Once again the author quotes some remarks made by other people.Paragraph 7 AnalysisIn this paragraph the author begins to cite examples in defense of netwriting. He first points out that there are actually some really good writings (gems) on the Internet.Paragraphs 8-9 AnalysisIn these two paragraphs the author discusses two reasons why there are some “gems” on the Internet: one is that the “Darwinian survival principle has started to prevail,” which means that only good writing will survive; and the other is the “collaborative” work in computer conferences, at which ideas are knocked against one another “until they spark.”Paragraph 10 AnalysisIn this paragraph the author explains one of the distinguishing features of netwriting (i.e. Netwriters freely lace their prose with strange acronyms and “smileys”), which, however, does not detract from its quality.Paragraph 11 AnalysisIn this paragraph the author mentions another merit of the Internet, that is, it allows thousands of common people to take up the craft of writing.Paragraphs 12-13 AnalysisThe last two paragraphs present a counter-argument to one of the criticisms against netwriting, that is, “It’s easy to make this stuff look foolish and trivial.” According toNielsen Hayden, “a lot of everyone’s daily life is foolish and trivial,” and besides, netwriting represents “for millions of people, a living, breathing life of letters.”II. Questions for ParagraphsParagraph 1: QuestionWhy does the author relate what happened to the telephone with what is happening to the computer?By doing so, the author seems to suggest that writing, which went out of style with the invention of the telephone, is experiencing an unexpected comeback with online letter writing.Paragraph 2: QuestionWhat is implied when the author says “... the media of McLuhan were supposed to render obsolete the medium of Shake speare ...”?The phrase “the media of McLuhan” refers to the new media that became popular in the 20th century such as radio, television and movies, while “the medium of Shakespeare” refers to the traditional way of “letter writing” in the main clause of t his sentence. The sentence implies that, instead of being rendered obsolete, letter writing is experiencing the fastest development since the 18th century, though it is in the “online” form.Paragraph 3: Questions1. What does the author mean when he say s “... E-mail and computer conferencing is teaching an entire generation about the flexibility and utility of prose”?The author means that E-mail and computer conferencing demonstrate to a whole generation of people that the language we use does not have to be always so formal as in traditional letter writing; it can be used in a casual manner while still serving its purpose effectively.2. What does the author wish to convey in his comparison between netwriting with “scribblers’ compacts,” Mark Twain’s d iscovery of new journalism in San Francisco, Revolutionary War pamphleteers and the achievements of the Elizabethan era? And what does he find with his comparison?The author uses these examples to tell the readers that there are some similarities between netwriting and those experimental, even innovative writing modes in history, and so netwriting could be regarded as a kind of renaissance. But he finds in thiscomparison that much of netwriting is awfully bad — sloppy, meandering, ungrammatical, poorly spelled, badly structured and even content free.Paragraphs 5-6: QuestionWhat are the reasons that could account for the poor quality of netwriting?First, things (such as E-mail) written on the Internet have little literary value, so they’re not meant to be permanent. They just come and go and no one will ever remember them, once they are read. Second, many online postings are produced under great time pressure. Third, there is actually no threshold on the Internet, so that anyone can log on and send out what they have written.Paragraph 7: QuestionWhat is implied in the sentence “... when I met Mark I was dressed as the Canterbury Tales. Rather difficult to do as you might suspect, but I wanted to make a certain impression”?In the sentence “Mark” refe rs to the fellow network scribbler mentioned in the preceding sentence. “I was dressed as the Canterbury Tales” means “I was dressed in a very old-fashioned medieval style.” The whole sentence suggests that Green wanted to attract attention.Paragraph 8: Questions1. What are the criteria for good writing favored by the Internet?Good writing on the Net should be clear, witty and brief. Units of thought are to be found in short paragraphs, bulleted lists and one-liners.2. Why does the author say that “a Darwinian survival principle has started to prevail”? There are three reasons that are assumed to be responsible for the poor quality of netwriting. The first has to do with the nature of the writing which is regarded as “written speech” and is of little value. The second has to do with the fact that some netwriting such as E-mail is written in a hurry. The last reason is the low barriers to entry of the online world, which is responsible for much second-rate fiction and poetry that would otherwise not have been exposed to the readers.Paragraph 10: Questions1. What does the author mean by “... what works on the computer networks isn’tnecessarily what works on paper”?The author is referring to different features between netwriting and published papers. Netwriters can freely lace their writings with strange acronyms and “smileys,” while the published prose copied onto bulletin boards from books and magazines often seems long-winded and phony.2. What are the differences between the work of netwriters and that of professional writers?Netwriting is usually terser than professional writing, filled with short paragraphs, bulleted lists and one-liners. Netwriters can freely lace their writings with strange acronyms and “smileys,” the key stroke combinations used to convey various facial expressions; while the polished prose of professional writers often seems long-winded and phony when it is copied onto bulletin boards from books and magazines.Paragraph 12: QuestionHow does the author justify the legitimacy of netwritng?The author contends that netwriting may seem foolish and trivial, but most people’s lives are foolish and trivial. For millions of people those networks represent a living, breathing life of letters.Paragraph 13: QuestionWho does “the Bard” refer to in the last sentence? And how does it differ from the “Bards” in the title?In the last sentence “the Bard” (with capital B) refers to William Shakespeare, while “Bards” in the title refers to the people who write on the Internet. It appears t hat the author suggests by the title that the medium of netwriting, like that of Shakespeare, is perfectly justifiable. The medium changes with the time, so does the “bard.”III. Language Work of ParagraphsParagraph 1“One of the unintended side effect s of the invention of the telephone was that writing went out of style.”Paraphrase: When telephone was invented, no one realized at that time that it would outdate writing.unintended adj. not deliberate or plannede.g. an unintended slightThe group argues that many of the proposed reforms will have unintendedconsequences.“... and gave their fingers — and sometimes their mind —a rest.” Explanation: ... and preferred to use the telephone rather than the pen (in a humorous way).Paragraph 2“Which makes what’s happening on the computer networks all the more startling.”Explanation: Here “which” is a relative pronoun referring to the situation mentioned in the preceding paragraph. The sentence would be grammatically acceptable if it were restructure d as “All this makes what’s happening on the computer networks all the more startling.”“Just when the media of McLuhan were supposed to render obsolete the medium of Shakespeare, the online world is experiencing the greatest boom in letter writing since the 18th century.”Paraphrase: Contrary to what is expected, at a time when the media of McLuhan (e.g. television) were supposed to make the medium of Shakespeare (i.e. letter writing) out of date, letter writing on the Internet is enjoying the greatest development since the 18th century.render vt. to make someone or something be or become somethingto express, show, or perform something in a particular waye.g. His rudeness rendered me speechless.The singers rendered the song with enthusiasm.obsolescent adj. (cf. obsolete) becoming replaced by something newer and more effectivee.g. The amateur movie gauges of 8 mm, Super 8 and 9.5 mm are obsolescent.Much of our existing military hardware is obsolescent.Note: Things that are obsolete are out of date or no longer in general use. Things that are obsolescent are fading from general use and soon to become obsolete.boom n. an increase in the activity of a particular industry or part of a country's economye.g. This year has seen a boom in book sales.The insurance business suffered from a vicious cycle of boom and bust.Paragraph 3David Sewell ... likens netwriting to the literary scene Mark Twaindiscovered in San Francisco in the 1860s, “when people were reinventing journalism by grafting it onto the tall-tale folk tradition.”Paraphrase: David Sewell compares netwriting to the literary scene Mark Twain discovered in San Francisco in the 1860s, “when journalists were fond of writing news reports in the traditional tall-tale style.”liken vt. l iken someone/something to somethingto say that someone or something is similar to someone or something elsee.g. She likened the experience to sinking into a warm bath.Our small company can be likened to a big, happy family.reinvent vt. to change something that already exists and give it a different form or purposee.g. The story of Romeo and Juliet was reinvented as a Los Angeles gangster movie.He’s one of those sportsmen who reinvent themselves as TV presenters.graft vt. to add something and make it become a part of another thinge.g. A piece of skin was removed from her leg and grafted onto her face.The management tried unsuccessfully to graft new working methods onto theexisting ways of doing things.the tall-tale folk tradition n. the tradition in which people tell a story or a tale in a fanciful and exaggerated manner.A tall tale/story is a tale or a story that is hard to believe, because it is so exaggerated or unlikely. Here the author refers to a unique kind of journalism in which people report news in much the same way as people tell tall tales.pamphleteers n. Though most pamphleteers were not professional writers, they were passionate advocators of North American independence.Paragraph 4“For it can be very bad indeed: sloppy, mean dering, puerile, ungrammatical, poorly spelled, badly structured and at times virtually content free.”Paraphrase:For it can be very bad indeed: careless, pointless, childish, with many grammatical and spelling mistakes, and structural mistakes, and sometimes it has no real message or meaning.sloppy adj. done in a very careless waye.g. Spelling mistakes always look sloppy in a formal letter.Another sloppy pass like that might lose them the whole match.meander vi. a river or road that meanders follows a path with a lot of turns and curves to talk or write for a long time, changing subjects or ideas, so that peoplebecome bored or confusede.g. a meandering rivera long meandering speechThe film meanders along with no particular story line.puerile adj. (n. puerility) like a silly young person, or like something they would say or doe.g. He can be very puerile when he’s had a couple of drinks.Some of your colleagues have been complaining about your puerility in the office.“I THINK METALLICA IZ REEL KOOL DOOD!1!!!”— This is an example of sloppy and vacuous writing on the Internet.Paragraph 5“... Gerard Van der Leun ... has emerged as one of the preeminent stylists on the Net.”Paraphrase:... Gerard Van der Leun ... has become known as one of the most important stylists on the Internet.emerge vi. to come out of something or out from behind somethingto become knowne.g. The facts behind the scandal are sure to emerge eventually.She’s the most exciting British singer to emerge on the pop s cene for a decade.preeminent adj. (n. preeminence) better or more important than anyone or anything else in a particular activitye.g. She is the preeminent authority in her subject.His preeminence in his subject is internationally recognized.“They’re not to have and hold; they’re to fire and forget.”Paraphrase:They are of little literary value, so people write them, send them and forget them.“Many online postings are composed ‘live’ with the clock ticking ...”Paraphrase: Many online postings are written spontaneously, with no preparation within a very short time ... (i.e. They are improvised under great time pressure.)Paragraph 6“... it takes a hell of a lot of work to get published, which naturally weeds out a lot of the garbage.”Paraphrase:... if you want to have your book published, you need to do a lot of work. In this way a lot of untalented writers are prevented from publishing bad works.a hell of a lot of: (used as an emphasizer) a great deal ofhell n. used for saying that something happens that makes people angry or upset, and they start fighting or arguinge.g. Oh hell, I’ve forgotten my key!What the hell was that noise?weed n. someone who is thin and weakweed out: to remove a person or thing that is not suitable or good enough, especially from a group or collectione.g. He looks like a real weed in those shorts.weed out dandelionsweed out unqualified applicantsa hell of a lot of: (used as an emphasizer) a great deal ofhell n. used for saying that something happens that makes people angry or upset, and they start fighting or arguinge.g. Oh hell, I’ve forgotten my key!What the hell was that noise?Paragraph 7ream n. a large quantity of somethinge.g. I ordered three reams of the best typing paperShe’s written reams of poetry.gem n.a beautiful expensive stone that is used to make jewellerysomeone who is special in some way, especially because they are useful or helpfule.g. a little gem of a bookShe inherited $20,000 in gold and gems.You’ve been an absolute gem —I couldn’t have managed without your help.“But even among the reams of bad poetry, gems are to be found.” Paraphrase:Even though most of the poems are of bad quality, there are still some really good ones.“He did, and blew them all away.”Paraphrase:He did publish his poems, and overwhelmed them all.Here “to blow someone away” is an informal expression which means “to makesomeone feel very surprised, especially about something they like or admire.” Italso means “to defeat someone completely.”blow away: to make someone feel very surprised, especially about something they like or admire; to defeat someone completelye.g. The ending will blow you away.That concert blew me away.I’m gonna blow him away when I catch up with him.Paragraph 8“... a Darwinian survival principle has started to prevail.”Paraphrase :... Darwin’s survival principle has started to come into effect / gain influence or control.prevail vi. to be the strongest influence or element in a situationto exist at a particular time or in a particular situatione.g. I am sure that common sense will prevail in the end.This attitude still prevails among the middle classes.The town is kept cool by the prevailing westerly winds.He was eventually prevailed upon to accept the appointment.Paragraph 9“... writers compose in a kind of collaborative heat, knocking ideas against one another until they spark.”Paraphrase:... writers work in a kind of cooperation by engaging themselves in heated discussions and arguments about different ideas until they come up with really brilliant ones.collaborate vi. (collaboration n.) to work with someone in order to produce somethinge.g. Two writers collaborated on the script for the film.A German company collaborated with a Swiss firm to develop the product.The two playwrights worked in close collaboration on the script.caliber n. (American English) qualitye.g. The department has the caliber of staff to make the project work.a film with a high-caliber castthe New York Times and the Wall Street Journaltwo high-quality newspapers. Here the author is trying to emphasize that there aresome gems on the Internet which major newspapers have printed excerpts from.Paragraph 10“Unless they adjust to the new medium, professional writers can c ome across as self-important blowhards in debates with more nimble networkers.”Paraphrase:If they do not adjust themselves to the medium of netwriting, they can make themselves look conceited and self-important in online debates with more quick-witted and flexible networkers.comes across:If someone or what they are saying comes across in a particular way, they make that impression on people who meet them or are listening to them.e.g. She comes across really well on television.What comes across in his later poetry is a great sense of sadness.comes across:If someone or what they are saying comes across in a particular way, they make that impression on people who meet them or are listening to them.e.g. She comes across really well on television.What comes across in his later poetry is a great sense of sadness.blowhard n. someone who talks too much about themselves or the things they have achievede.g. Stop being such a blowhard! Do you have to talk about yourself the whole time? nimble adj. able to move quickly and easilye.g. nimble witsHis nimble mind calculated the answer before I could key the numbers into mycomputer.their blue-blooded peers: Blue-blooded people are usually members of royal or noble families. Here the author refers to well-educated children.Paragraph 11inherently democratizing: Here the author suggests that writing online (the technology) provides equal opportunities for everyone who wishes to write.senior citizens: a euphemistic term for “old people”computer geeks: (American slang) Here the phrase refers to computer enthusiasts.“Not only has it enfranchised thousands of would-be writers who otherwise might never have taken up the craft, but it has also thrown together classes of people who hadn’t had much direct contact before…”Paraphrase : It has not only encouraged thousands of potential writers who, without the introduction of the Net, might never have become writers, but has also brought together people of different social backgrounds who hadn’t had much communicati on before ...enfranchise vt. to give someone the right to votee.g. Women in Britain were first enfranchised in 1918.take up: to accept an offer or a challenge (=an offer to fight or compete) that someone has made to you to start doing something regularly as a habit, job, or intereste.g. He’s taken up the post of supervisor.She’s just taken up cycling to work.I paused and my friend took the story up for a while.We’re not very good at French, we only took it up recently.craft vt. (craftsmanship n.) to make or produce something skillfullye.g. craft workersliterary crafttraditional craftsThe jewelry showed exquisite craftsmanshipParagraph 12“After all, a lot of everyone’s daily life is foolish and trivial.”Paraphrase:In spite of everything, our daily life consists mostly of foolish and insignificant things.travail adj. (triviality n. trivia n. trivialise vt.) not very important, serious, or valuablee.g. Getting computers to understand human language is not a trivial problem.I’m a busy man —don’t bother me with trivialities.I’m fascinated by the trivia of everyday life.I don’t want to trivialise the problem, but I do think there are more importantmatters to discuss.“I mean, really, smileys?”Paraphrase: I mean, smileys are really foolish and trivial.“Housewives in Des Moines who log on as VIXEN?”Paraphrase: Can we take seriously the writing of housewives who live in a less-known place such as Des Moines and use VIXEN as their user name?“But it would be a mistake ... to underestimat e the effect a lifetime of。

A First Course in ProbabilityIntroductionProbability theory is a branch of mathematics that deals with the study of randomness and uncertainty. It provides a framework for understanding and quantifying uncertainties in various fields, ranging from finance and economics to engineering and science. A First Course in Probability aims to introduce the fundamental concepts and techniques of probability theory and provide a solid foundation for further study in the subject.Basic Probability TheorySample Space and EventsIn probability theory, we start by defining a sample space, denoted by Ω, which is the set of all possible outcomes of an experiment. An event, denoted by A, is a subset of the sample space. The probability of an event is a real number between 0 and 1, representing the likelihood of that event occurring.The Calculus of ProbabilityThe basic operations of probability include union, intersection, and complement. Given two events A and B, the union of A and B, denoted by A ∪ B, consists of a ll outcomes that belong to either A or B. The intersection of A and B, denoted by A ∩ B, consists of all outcomes that belong to bothA and B. The complement of an event A, denoted by A’, consists of all outcomes that do not belong to A.The probability of the union of two events is given by the sum of their individual probabilities minus the probability of their intersection:P(A ∪ B) = P(A) + P(B) - P(A ∩ B)Conditional ProbabilityConditional probability measures the likelihood of an event A occurring given that another event B has already occurred. It is denoted by P(A|B) and is defined as:P(A|B) = P(A ∩ B) / P(B), where P(B) > 0IndependenceTwo events A and B are said to be independent if the occurrence of one event does not affect the probability of the other event. Mathematically, two events are independent if and only if:P(A ∩ B) = P(A) * P(B)Random VariablesA random variable is a function that assigns a real number to each outcome in the sample space. It provides a way to quantify the uncertainty associated with an experiment. Random variables can be discrete or continuous, depending onwhether they take on a countable or uncountable number of values, respectively.Probability DistributionsDiscrete Probability DistributionsIn the case of discrete random variables, the probability distribution can be defined by a probability mass function (PMF), which gives the probability of each possible value of the random variable. The PMF satisfies two properties: it must be non-negative for all values of the random variable, and the sum of the probabilities must equal 1.Examples of discrete probability distributions include the Bernoulli distribution, the binomial distribution, and the Poisson distribution.Continuous Probability DistributionsFor continuous random variables, the probability distribution is defined by a probability density function (PDF), which specifies the relative likelihood of the random variable taking on different values. The PDF must be non-negative, and the total area under the curve must equal 1.Examples of continuous probability distributions include the normal distribution, the exponential distribution, and the uniform distribution.Expectation and VarianceExpectationThe expectation of a random variable, denoted by E(X), is a measure of its average value. For discrete random variables, the expectation is calculated by summing the product of each possible value and its corresponding probability. For continuous random variables, the expectation is calculated by integrating the product of each value and its corresponding density over the entire range of values.VarianceThe variance of a random variable, denoted by Var(X), measures the spread or dispersion of its probability distribution. It quantifies how far the values of the random variable deviate from its expectation. The variance is calculated by taking the expectation of the squared difference between each value and the expectation.Central Limit TheoremThe central limit theorem states that the sum or average of a large number of independent and identically distributed random variables will be approximately normally distributed, regardless of the shape of the original distribution. This theorem has wide-ranging applications in statistics and allows us to make inferences about population parameters based on sample data.ConclusionA First Course in Probability provides a solid foundation in the fundamental concepts and techniques of probability theory. It covers basic probability theory, probability distributions, expectation and variance, and the central limit theorem. This course serves as a starting point for further study in the field of probability and its applications.。

Scaling up The Training of Deep CNNs for HumanAction RecognitionM.Sai RajeswarIndian Institute of Technology Delhi Email:rajsai24@ A.Ravi Sankar,Vineeth N.BalasubramanianIndian Institute of Technology HyderabadEmail:cs11m03@iith.ac.in,vineethnb@iith.ac.inC.D.SudheerIBM Research,New DelhiEmail:sudheer.chunduri@Abstract—Convolutional deep neural networks(CNNs)has been shown to perform well in difficult learning tasks such as object recognition.They are gaining huge importance in recent times but are computationally intensive.Typically trained on massive datasets,two-dimensional CNNs are used for image classification and recognition purposes and consume huge com-putational time.For applications like human action recognition involving video inputs,their3D counterparts termed as3D convolutional neural networks(3D-CNNs)are employed.Scaling up the computations to support large datasets and accelerating the training on these models for high performance has been the need of the hour especially in3D deep learning models since the extended connectivity of CNN in the time domain takes huge time for training the model.Also there is a need to look at the model parameters and hyper parameters that determine both the computational performance as well as the accuracy of the deep neural network.Accelerators such as Graphics Processing Units(GPUs)and multi-cores provide a means for speeding up the training of CNNs and achieve higher performance by parallelizing the training of these models by taking advantage of data and model parallelism.In this work we use multi-core CPUs and GPUs to scale-up the training of3D-CNNs.We achieve a faster implementation as well as report how various network parameters affect the performance of the model thereby providing recommendations on initializing the values of the same.The code scales up well on multi-cores and GPUs,with a speedup of10x on CPUs and achieves almost12x on GPUs compared to the serial version.Our work infers that3D-CNN code scales up best on CPUs when the convolution step is implemented with a highly parallel FFT based approach,thereby achieving the performance comparable to GPUs using OpenMP.I.I NTRODUCTIONConvolutional neural networks are effective and powerful deep learning models in the wide area of artificial intelligence. Their structure and design is inspired from real human brain model and are currently used extensively in image object recognition[19],automatic speech recognition[8],video tag-ging[18]and processing,which are some of the most important applications of machine learning.Important reasons behind thier extensive application and efective results being scalability of the networks to huge labelled datasets and parameters involved in learning process and secondly,automated process of learning interpretable and powerful feature sets from the raw data.Infact,it is a well proven fact that when CNNs are trained using adequate regularization,they can attain very high performance without utilising handcrafted features on visual object classification[6].Encouraged by these facts they are applied heavily in video classificaion problems in a slightly extended form from what has been used in images or speech data,generally known as3D Convolutional Neural Networks(3D-CNN).Human action recognition[18]in real world environment is one of the main applications of3D-CNNs where in they are employed to classify a set of human actions from a large-scale video data.From the computational point of view,3D-CNNs are very expensive in terms of extremely overwhelming high periods of training time[9]of neural networks to optimize thousands of parameters that define the network model in an effective manner.As the model needs to train over by processing several frames of video rather than a single image at a time,there is an immediate need in case of deep learning models such as3D-CNNs to be effectively trained using less computational time and practically tractable by accelerating things up.One of the efforts in this direction has been the distributed computing environment which was known as DistBeleif[4]where in 16000CPU cores are used for training a network model taking just few days,but seldom these resources are available to researchers.Therefore,there is a dire need to leverage inex-pensive and efficient computing capability in the form of GPUs and multi-cores,wherein the underlying model itself architec-tured in a way to adopt,utilise and scale up well with these resources.To handle the above specified requirements and mitigate the disadvantages,parallel frameworks are essential in getting the neural network model train faster,bigger and easier by deviating from the plain implementation of the3D-CNNs. Theano[16]a deep learning library provides a way for the efficient utilisation of such CUDA capable devices and multi-core CPUs,while retaining the original functionality of the deep learning model.In order to achieve higher performance there are two main alternatives in parallelizing the training of 3D-CNNs which are in general applicable for any CNN model. First being,different threads or nodes working across different training data which is often called data parallelism and second, the computations on a single data can be parallelized where different threads or nodes holding on to different parts of the model share the work,often termed as model parallelism.In this work we leverage on both these techniques for training 3D-CNNs that utilise GPUs and multi-cores to achieve very high scalability in the computational capability of these neural networks.Parameters and Hyper-parameters involved in the training of3D-CNNs or in general CNNs include batch size,number of feature maps,pooling parameters,learning rate etc.They are crucial in determining both the training accuracies as well as affect the computational performance of the network. In this work we provide a recommendation based approach for achieving decent speedup taking into the fact that speed2015 IEEE International Parallel and Distributed Processing Symposium Workshops 2015 IEEE International Parallel and Distributed Processing Symposium Workshopdoesnot counter act on the over all performance of the model. The parameter study is again out of various experiments on a 16-core CPU using OpenMP.Considering that the computational speed is a drawback of CNNs,there has been quite some work done on improving their performance.Patrice Simard et al[7]has show in their work on document processing that integrating CNNs with BLAS libraries produces2.4x to3x speedup and using GPUs the speedup can be upto3.1x to4.1x,wherein their GPU implementation is based on pixel shader approach[7].A High performance approach for image classification was used by Schmiduber et al[10]wherein a2-3x speedup was reported by just switching from GTX285to GTX480as the latter has the advantage of a R/W L2global cache for device memory.More recent work on parallelizing CNNs by Alex Krizhevsky[17] is based on the algorithm that implements data prallelism in convolutional layers and model parallelism in fully connected layers.This approach gave comparable results on large image datasets.Krizhevsky’s open source platform cuda-convnet2can be employed to view the results,some other platforms for efficient CNN implementation includes Caffe by UC Berkely and Torch7.All the before-specified optmizations were done for CNNs that accept2D input data such as images,very little work is done in optimizing the3D-CNNs and none of the platforms mentioned provide an efficient implementation of3D-CNNs. This paper presents a novel high performance optimization of 3D-CNNs for video action recognition.We demonstrate the capability to train a3D-CNN model involving thousands of parameters or weights,scaling12x times more on GPU and 10x more on muti-cores which were the best results obtained by employing three novel optimization approaches to speedup 3D-CNNs that are outlined later in the paragraph.The rest of this paper is organised as follows.Breif description of3D-CNN architecture is presented in section2. In section3,4and5we discuss the details of three major optimization employed in our work are presented,firstly a 3D Fast Fourier Transform(FFT)based convolution imple-mentation and loop unrolloing of pooling step which converts the computation in each convolution layer into vector SIMD product operations followed by MKL based performance opti-mizations and lastly porting the code to multi-cores and GPUs using OpenMP and CUDA for higher performance.The results of experiments which include emphasis on network parameters are reported in section6.Concluding remarks and the future work are provided in section7.II.3D-CNN A RCHITECTURECNNs are hierarchical and structured neural nets comprised of a convolutional layer followed by subsampling layer and multiple such units stacked together.These are followed by one or more fully connected layers and a classification layer, a structure inspired by the neural cells present in the visual cortex of human brain[11].In applications concerned with2D data like images,convolutional layers compute2D convolution to obtain high-level features from the output feature maps of the layer on the right of them.The subsampling layer reduces the dimentionality of the features from convolutional layers over a local neighbourhood either using maxpooling,where the maximum element in a pooling window is taken or average pooling strategy wherein average value of the elements of the window is taken.3D-CNNs extend the same architecture as above except that now temporal direction is added to compute the feature maps. The architecture is outlined infigure1.Instead of just one image,a sequence of images is convolved with the3Dfilter. The architecure accepts input data as minibatches,one such model could consist a input of a sequence of nine images of size60by80convolved with5channels offilters of size4x7 x9separately and subsampling layer with the pooling window size3x3.The remaining layers are structured according to the output dimensions of the previous layers and thefinal output is passed to softmax/logisticlayer.Fig.1:3D-CNN ArchitectureIII.U NROLLING THE POOLING STEP The two time-critical steps in convolutional neural net-works are convolution and pooling routines.Every convolu-tional layer of a3D-CNN computes a discrete three dimen-sional convolution operation of a video frame with a3D filter,after which a non-linear transfer function is applied. Pooling is an important step in the subsampling layer of CNNs that reduces the feature map resolution and achieves spatial invariance.Naive straight forward implementation of pooling step in video classification is achieved using four nested for-loops,the outer one ranging across different kernels,second ranging across images in a frame,the third with in an image across the row,the last with in an image across the columns. The smaller thefilter size,higher are the jump instructions executed as a part of the code which makes this approach even more inefficient.One optimization could be removing the redundant for loops by vectorizing the operations and thereby utilising the SIMD instructions internally by the CPU.In this case the presence of special optimized hardware like MMX or SSE could be leveraged and leads to better performance.All the relevant arrays used are python numPy arrays which does utilise the SIMD instructions internally.Fast Fourier Transform(FFT)based convolution operations are mathematically proven to be faster and has inherent par-allel formulation.FFT transforms convolution operation into computing matrix multiplication,providing a scope for parallel implementation of matrix ing the cuda based FFT computations such as cuFFT and cuIFFT can leverage this parallelism.Even without utilising full power of GPUs wecould see that sheer implementation of FFT based convolutionhas given rise to1.85x speedup proving the theoretical factthat it is indeed faster.IV.M ATH K ERNEL L IBRARY AND MULTI-COREO PTIMIZATIONSConvolutional operation either direct or high dimensionalmatrix product based approaches can be optimized by efficientlinear algebraic vector operations,which essentially are used inthese approaches.Rather than slow,naive implementations oflinear algebra operations,tools that could grip the parallelismacross these computations yeilds better performance.IntelMath Kernel Library(MKL)are a powerful set of tools handdesigned to accelerate the application performance that involveintense vector math computations and harness the power ofmulti-core CPUs.Basic Linear Algebra Subroutines(BLAS)and Linar Algebra Package(LAPACK)and many other func-tionalities make MKL a source package for performance.Mostof the the data in the video classification applications are storedin Numpy vectors.Numpy when integrated with MKL mapsthe matrix and vector operations into highly efficient BLASand LAPACK routines as needed and boosts the performance.Utilizing the MKL libraries for3D-CNN code has reducedthe computation time to a good extent giving2.4x speedup.Ex-periments conducted on processors utilizing different numberof cores were also described in results section.Other advan-tages of MKL is that the calling sequences are standardizedso that programs that call BLAS will work on any computerthat has BLAS installed.V.3D-CNN ON M ULTI-CORES AND GPU WITH T HEANO GPUs and multi-cores have been applied successfully toincrease the performance resulsts of convolutional neural net-works on2-Dimensional data applications.We here proposea GPU based and multi-cores based implementations of3D-CNNs for video action recognition.A.OpenMP usage on multi-coresOpenMP pragmas are used to achieve parallelization onmulti-core CPU machines.Here again both convolution stepand subsampling step can be optimized by unrolling the outerloops with OpenMP pragmas.Though overhead might bean issue,as long as the performance doesnot degrade whilescaling to higher number of cores,the experiment can beemperically optimized.Infact,Theano supports OpenMP andprovides efficient way of leveraging multiple cores with itsinbuiltflags and utilizing the default environment variable OMP_NUM_THREADS.Various experiments on different para-menters for OpenMP were conducted and the same havebeen reported in the results section,which show that multi-core optimizations are promising.parallelism is leveraged withhighly parallel natured FFT based convolutions.Convolutiontheorem[13]states that convolution of any two functionsprecisely equals the product of Fourier transforms of thosefunctions which essentially boils the implementation down tocomputing FFT of the input data,perform product operationsand compute inverse FFT to get back the results.OpenMP isagain used with FFT algorithm to run parallely,wherein thethreads run the butterfly structured FFT framework leveraging the indepence present in the algorithm.Infact,the results are comparable to some of the best implementations on GPUs.B.GPU OptimizationsThe training of CNNs is comprised of hugefloating point operations and considerably less transfer of data in every training epoch,which makes it perfectly suitable for run-ning on the powerful graphics processing units.GPUs have a highly parallel architectures built for achieving maximum throughput,for example,NVIDIA GTX580has512cores with 1.5TFlops/sec forfloat32compared to Intel Core i7980,which has16cores with107GFlops/sec forfloat64.Efficiently Util-ising these powerful devices by leveraging on data and model parallelism could lead to high performance implementation of 3D-CNNs for video action ing mini-batches in3D-CNNs leads to data-parallelism where in multi-model replicas are launched for various mini-batches of the video data which leads to different worker threads working across the data simultaneously.In thefirst approach to parallelize on GPU,work for many CUDA threads is generated in a generic way.As an instance, inorder to add two matrices,the same addition operation is done on all the elements.Now these kind of operations allows a way to efficiently generate and distribute work for many CUDA threads.3D-CNNs are comprised of operations of these kinds,both in convolution layer and the subsampling layer of the model,where the output is downsampled by shifting the filter by more than one pixel each time.In the convolution layer,the convolution operation between the video frame and the3Dfilter is implemented using a series of2D convolutions through a constrained matrix multiplication operations which are normally called as tensor operations.Now,this allows a way to launch similar such operations across the kernel threads of GPU and compute multiple convolutions at the same time. Using the same approach GPU parallelism is leveraged also with FFT based convolutions.All the three steps involved can be parallelized by using CUDA implementation of FFT i.e. cuFFT,followed by simultaneous execution of tensor or matrix products with efficient CUDA kernel andfinally use cuIFFT for Inverse fourier transform.C.CUDA GEMM based optimizationNVIDIA CUDA provides a highly optimized implemen-tation of General Matrix Multiplication(GEMM)subroutines through cuBLAS library.GEMM(SGEMM/DGEMM)sin-gle/double precision matrix multiplication subroutines are heavily used in scientific computing and the same could be used for the3D-CNN model for convolution operation if convolution operation could be converted into product of small matrices.It is infact time consuming to optimize completely a given function to port it to GPU.NVIDIA did that for GEMM operations[1].These optimizations provided through GEMM are comparable and sometimes supersede the optimized Math Kernel Libraries of Intel parallel implementations.One such productive way of transforming convolution op-eration into matrix product operation is Toeplitz matrices[12]. Toeplitz matrices are diagonal-constant matrices,wherein ev-ery element on a diagonal entry from top-left to right bottom are constant.A convolution operation can be transformedconstructively into matrix multiplication,where one of the inputs is constructed to a toeplitz matrix.Unlike FFT,there is no overhead of taking FFT on both inputs and finally converting them back using IFFT and doesnot involve much memory tranferring issues that can slow things down.For each input frame in the mini-batch of 3D-CNN,a toeplitz matrix is created in a cuda kernel follwed by GEMM call to multiply matrix with the appropriate filter.This implementation has given the best performance in all our experiments as explained later.One limiting case for this approach is the requirement of extra copy of memory when converting into toeplitz matrix,other than that it turns out to be one of the most effective implementation of convolution exended to a high dimensional matrix operations in 3D-CNNs.VI.R ESULTSThe plain vanilla functionality of 3D-CNN for video ac-tion recognition implemented in python was taking roughly 200mins for 10epochs,which is very high considering the fact that CNN needs very large number of epochs to train.A series of optimizations were introduced in a progressive manner in various experiments conducted.At the end,the computational time was drastically brought down.This work is the first to report optimal results for speedingup action recognition using 3D-CNN.A.Experimental setupThe code was run on Stampede Supercomputer(TACC)which is comprised over 100,000cores with FDR infini-band interconnect network wherein every core works at 2.7GHz(Intel Sandy Bridge,24MB L3cache)including 32GB memory for a 16-core node.For GPU experiments NVIDIA Tesla K20card consisiting of 6GB memory(both global/device)with 448cores was utilized for GPU based implementation.The code employed in this performance study was compiled with Intel compiler version(14.0)using opti-mization flags -O3,-fast and -openmp.Two standard datasets were used with varying sizes-KTH[15]with six action cate-gories and UCF11[14]sports dataset comprising thousands of Youtube clippings with 11action categories,these datasets are commonly used in the action recognition community.Walking and running action frames of KTH dataset along with the respective features learnt by the model are shown in Fig.2.UCF11is almost double the size of KTH in terms of dataset size.Experiments were carried on both datasets and results have been summarizedaccordingly.Fig.2:Walking and running actions from KTH-dataset andon the right are the corresponding features learntB.CPU optimizationsSome of the processor optimizations included loop un-rolling for the costly pooling step and MKL based opti-mizations to speedup the convolutions.There is redundancyin the nested looping structure in the direct implementation of pooling.This is optimized by loop unrolling approach,wherein concurrency is leveraged by vectorized operations.As expected,the runtime has further decreased to half from the original due to the SIMD nature of numpy arrays.Also,a plain implementation of FFT based convolution instead of direct implementation of 3D convolution further gave a 1.7x speedup.Integrating optimized MKL with 3D-CNN implementation has doubled the computational speed as shown in Table III.TABLE I:multi-core optimization timings for 10-epochsTHREADS Timing for KTH(sec)Timing for UCF111638.693899.192288.038412.84170.15294.78112.42204.416136.23140.8OpenMP was used for accelerating the performance and computational time is reported in Table.I for various number of threads enabled through environment variable OMP_NUM_THREADS .For KTH dataset the best result is observed for 8threads and for 16threads there is a negative improvement as shown in Fig.3,typically this happens when the overhead due to parallelization,which increases with the number of cores becomes larger than the gain in computation time.Its a trade-off again.For the larger UCF11dataset 16threads is still an improvement.1)FFT based Convolution using OpenMP:As a second optmization,knowing that convolution step is one of the main time consuming operation,a 3d FFT based convolution is im-plemented.This boosted the speedup given theparallel natured FFT operation and matrix multiplication.FFT implementation with OpenMP for various thread-numbers was plotted in Fig.4,with 16threads the speedup is almost double compared to the plain convolution implementation.When trained on a larger dataset with larger mini-batch size FFT did scale-up well as shown in the results.Fig.3:Scaling with OpenMPC.GPU optimizationsA lot of experiments were conducted on the aforemen-tioned datasets using GPUs with various distinct and disjoint approaches and the results are tabulated as shown in TableII and III.Firstly,3d convolution has been implemented as a series of 2d convolutions as specified.FFT based convolutionFig.4:Scaling with OpenMP using FFTwas also implemented for GPUs but the speedup is not so great especially when the batch size is less andfilters are small mainly because of memory st optimization is a GEMM based approach on Toeplitz matrix transformation which on efficient parallelization gives the best results.The code scales well on larger datasets,as the datasize is doubled the speedup is accordingly scaled.For instance,the GPU implementation takes1.28mins on KTH dataset for10epochs, for UCF11dataset which is almost double the size of KTH the code takes2.51mins.The GEMM based approach takes 51seconds for10epochs and for the dataset that is double the size it takes1.38mins.So,there is a nice pattern seen scaling with the data.In all of the GPU based optimizations Theano library is used for the effective implementation of CNNs.Of all the implementations,it is the GPU based computations that gave the higher performance.For extremely large datasets,it would be promising to carryout the computations across many more GPUs with efficient communication.TABLE II:GPU optimization timings for KTHdataset(10-epochs)Optimization Timing(sec)SpeedupGEMM based5112X3D from2D987XFFT based498 1.5XMKL optimisation210 3.5XTABLE III:GPU optimization timings for UCF11(Larger)datasetOptimization Timing(sec)SpeedupGEMM based9213X3D from2D1526XFFT based589 1.8XMKL optimisation380 3.5XD.Study of architectural parameters on performanceStudy of computational performance with respect to various architectural parameters of CNN could be very important and lead to significant improvement regarding the performance of the model.Some of the parameters include batchsize,number of feature maps and pooling size.1)Batchsize:Data parallelism in3D-CNN is essentially due to batchsize,large batchsize provides more work to be divided across the threads and this is observed in Fig.5.But as the batchsize increases,the rate of error decrement across epochs falls low and the trade-off between computational efficiency and the qualitative efficiency of the network should be takencare.The plot also shows3D-FFT based convolution implementation using OpenMP.In both the cases,where convolution is implemented with and without FFT,experiments were carried on with16OMP threads since this was the optimum for UCF11dataset.TABLE IV:Effect of Batch-size on speedup for CPUbatch size Timing for UCF11(min)28.63 4.015 3.1010 2.8120 2.3450 2.2Fig.5:Batchsize vs time taken2)Number of Feature maps:As the number of feature maps are reduced across layers,the computational time de-creases,which is intuitive given that there are lesser computa-tions,but more feature maps would lead to better abstraction and higher accuracy.Table V shows the effect of change in the number of feature maps at thefirst layer,similar conclusions can be drawn when its increased or decreased at other layers. TABLE V:Effect of Number of Feature-maps on speedup Number of feature-maps Timing for UCF11(min)4 3.55 4.17(batch20)7.115(batch50)12.43)Pooling size:As the poolsize decreases,the compu-tational time increases.With the increase in poolsize,the performance increases in computational point of view,since for a larger poolsize parallelism can be exploited more for a single pooling operation.But as mentioned earlier,increasing the pooling size again decreases the rate of error decrement,where the trade-off needs to be taken care.Pool size is varied so that the output of the respective convolutional layer is divisible by pool size.Effect of pool size can be studied with varying batch size too,as shown in table VI,which again as before mentionedcan be chosen optimally.These numbers again depends on the size and quality of the underlying data.TABLE VI:Effect of Pooling-size on speedup for CPUpool size Timing for UCF11(min)2,2 4.83,3 2.685,5(batch20) 2.295,5(batch50) 2.36VII.C ONCLUSIONA comprehensive view of various optimizations for high performance implementation of3D-CNNs were proposed in-order to accelerate their training,which otherwise takes huge amount of time.Our main focus through out was to optimize the two time consuming operations observed after thorough profiling of the code,convolution and pooling.These opera-tions being a variation of matrix/tensor based operations,MKL and GEMM could be succesfully applied to implement faster versions,both on multi-core and GPUs.Also,studied how performance is affected by the network parameters,thereby striking a balance between speedup and overall accuracy by choosing optimum values for the parameters.We noticed a very good scaleup almost11x on CPUs with effecient implem-ntation of OpenMP enabled FFT based convolutions,when compared to naive implementation of3D-CNN.We would also like to extend the work by porting the application onto multi-node and multi-GPU platform with efficient communication across the devices using MPI which could definitely enhance the performance.R EFERENCES[1]Chetan Jhurani,Paul Mullowney.A GEMM interface and implemen-tation on NVIDIA GPUs for multiple small matrices Arxiv preprint arXiv:1304.7053,2013.[2] D.C.Cire an,U.Meier,J.Masci,L.M.Gambardella,and J.Schmidhu-ber.High-performance neural networks for visual object classification Arxiv preprint arXiv:1102.0183,2011.[3]Krizhevsky A.Parallelizing Convolutional Neural Networks.utorial ofIEEE Conference on Computer Vision and Pattern Recognition(CVPR 2014),(Columbus,Ohio,USA,June23-28,2014).2014.[4]Dean,J.,Corrado,G.S.,Monga,R.,et al,Ng,rge ScaleDistributed Deep Networks.Proceedings of the Neural Information Processing Systems(NIPS12(Lake Tahoe,Nevada,United States, December36,2012).Curran Associates,Inc,57Morehouse Lane,Red Hook,NY,2013,1223-1232.[5]Coates, A.,Huval, B.,Wang,T.,Wu, D.J.,Ng, A.Y.learningwith COTS HPC systems In Proceedings of the30th International Conference on Machine Learning(ICML13),(Atlanta,Georgia,USA, June1621,2013).JMLR:W and CP volume28(3),2013,1337-1345 [6]Krizhevsky,A.,Sutskever,I.,and Hinton,G.E.Classification with DeepConvolutional Neural Networks.Proceedings of the Neural Information Processing Systems(NIPS12),(Lake Tahoe,Nevada,United States, December36,2012).Curran Associates,Inc,57Morehouse Lane,Red Hook,NY,2013,1097-1106.[7]Kumar Chellapilla,Sidd Puri and Patrice Simard.High PerformanceConvolutional Neural Networks for Document Processing Journal of the American Statistical Association,pages1200–1224,1995.[8]Hinton,G.,Deng,L.,Yu,D.,et al,Kingsbury,B.Deep Neural Networksfor Acoustic Modeling in Speech Recognition.Signal Processing Magazine,29,6(November.2012),82-97.[9]M.Baccouche,F.Mamalet,C.Wolf,C.Garcia,and A.Baskurt.Se-quential deep learning for human action recognition.Human Behavior Understanding,pages2939.Springer,2011.2,3[10]Dan C.Ciresan,Ueli Meier,Jonathan Masci,Luca M.Gambardella,Jurgen Schmidhuber.Flexible,High Performance Convolutional Neural Networks for Image Classification Proceedings of the Twenty-Second International Joint Conference on Artificial Intelligence.Chapman and Hall.,New York,USA,2008.[11]Davoud Nouri,Hamid Karimi Capability of Convolutional NeuralNetworks to Represent Human Visual System.2nd Basic and Clinical Neuroscience Congress,Tehran,Iran[12]Dave Meko.Discrete Convolution using Toeplitz matrices https://en./wiki/Toeplitz matrix#Discrete convolution.Accessed: 2014-07-19.[13]Convolution theorem /wiki/Convolutiontheorem.Accessed:2014-07-19.[14]UCF11Dataset /data/UCF YouTube Action.php.Accessed:2014-07-19.[15]KTH Dataset http://www.nada.kth.se/cvap/actions/.Accessed:2014-07-19.[16]Bergstra,J.,Breuleux,O.,Bastien,F.,et al,Bengio,Y.Theano:A CPUand GPU Math Expression Compiler.Proceedings of the Python for Scientific Computing Conference(SciPy2010),(Austin,Texas,USA.June30-July3,2010).[17]Alex Krizhevsky.One weird trick for parallelizing convolutional neuralnetworks Unpublished manuscript.2014[18]Shuiwang Ji,Wei Xu,Ming Yang,Member,IEEE,and Kai Yu,Member,IEEE.3D Convolutional Neural Networks for Human Action Recognition.IEEE Transactions on Pattern Analysis and Machine Intelligence,vol.35,no.1,January2013[19]Yongqiang Zou,Xing Jin,Yi Li,Zhimao Guo,Eryu Wang,Bin XiaoMariana:Tencent Deep Learning Platform and its Applications.Joseph Henry Press,1997.。

Reading beyond the linesA typical way to develop perspective is to consider issues from different vantage points. A less common - but perhaps more valuable - way to develop perspective is to expose yourself to different ideas.One way to do this is to read books that are at the fringes of your "typical" interest area - and then apply those concepts in your area of expertise and in your dealings with others. Here are some recent books that I have read, along with the interesting lessons that they taught:Working the Room (Nick Morgan): If it is important enough to get up in front of a group and say something, then it is important enough to rehearse.The Trusted Advisor (David Maister): Personal credibility and relationships are what develop long-term relationships with clients - not price.Chaos (James Gleick): The seminal book on the origins and uses of Chaos theory; The biggest advances in science are often discovered by those whose training started in a different field. Conventional wisdom sometimes prevents progress.Freakonomics (Steven Levitt/Stephen Dubner): Personal motivation and incentive are often at the heart of behavior.Growing an Engaged Church (Albert Winseman): The connection between personal engagement and the development of a community of interest.Getting to Yes (Roger Fisher /William Ury): Wear the other person's shoes: helping others get what they want is the first step to getting what you want.Linked (Albert-Laszlo Barabasi): It's not the "nodes", but the connections between them that make the network.Carpe Diem: How to become a latin lover (Harry Mount): This book is a description and primer on Latin; it illustrates what can be learned from a culture from its language. (And conversely, how hard it is to learn about a culture if you don't know its language.)Death Sentences (Don Watson): This book rails against the use of weasel words and management-speak.Godel Escher Bach, an Eternal Golden Braid (Douglas Hofstadter): The intimate, and sometimes surprising, connection between philosophy, art, and music.Leaving the Mother Ship and Personal Balance Sheet: More so than reading, writing about a subject in depth forces you to think deeply about it.What's on your list - and what did you learn?A good reader is one who can read beyond the lines, seeing ideas undeclared through the words, and one who can bridge the gap between the obvious and the suggested, thus getting much more information. Let’s look at the following sentence:John Henderson was driving home late last night from a tiring business trip.As you might have realized, this is the opening sentence of the story THE HITCHHIKER (Reading Practice 10, Book I). It tells the reader whom the story is about—John Henderson, and when the event takes place—late one night. If we read carefully and thoughtfully, we can also infer from this sentence that:1. John Henderson may be a businessman because he has just finished a “business trip”;2. John Henderson must be very tired because the trip is “tiring”;3. John Henderson must be very eager to get home because he is driving “late” at night right after this tiring trip.Making reasonable guesses while reading will always help us achieve a better understanding. But we should keep in mind that judgments must be made with care and supported by proof. In order to make reasonable judgments, we should1. read and think beyond the printed words;2. examine the information given in the passage;3. determine the writ er’s reason for his or her choice of words.。

a rX iv:mat h /311254v2[mat h.PR]6Apr25The Annals of Probability 2004,Vol.32,No.4,2857–2883DOI:10.1214/009117904000000568c Institute of Mathematical Statistics ,2004THE BROWNIAN WEB:CHARACTERIZATION AND CONVERGENCE By L.R.G.Fontes 1,M.Isopi 2, C.M.Newman 3and K.Ravishankar 4Universidade de S˜a o Paulo,Universit`a di Roma,New York University and SUNY College at New Paltz The Brownian web (BW)is the random network formally con-sisting of the paths of coalescing one-dimensional Brownian motions starting from every space-time point in R ×R .We extend the earlier work of Arratia and of T´o th and Werner by providing a new char-acterization which is then used to obtain convergence results for the BW distribution,including convergence of the system of all coalescing random walks to the BW under diffusive space-time scaling.1.Introduction.In this paper,we present a number of results concern-ing the characterization of and convergence to a striking stochastic object called the Brownian web (BW).Several of the main results were previously announced,with sketches of the proofs,in [13].Roughly speaking,the BW is the collection of graphs of coalescing one-dimensional Brownian motions (with unit diffusion constant and zero drift)starting from all possible starting points in one plus one-dimensional (con-tinuous)space-time.This object was originally studied more than twenty years ago by Arratia [5],motivated by asymptotics of one-dimensional voter models,and then about five years ago by T´o th and Werner [26],motivated by the problem of constructing continuum “self-repelling motions.”Our own interest in this object arose because of its relevance to “aging”in statistical physics models of one-dimensional coarsening [14,15]—which returns us to2FONTES,ISOPI,NEWMAN AND RAVISHANKARArratia’s original context of voter models,or equivalently coalescing random walks in one dimension.This motivation leads to our primary concern with weak convergence results,which in turn requires a careful choice of space for the BW so as to obtain useful characterization criteria for its distribution. We remark that there are two questions we do not address in this paper that are worthy of consideration.Thefirst is whether our convergence results might play some role in studying the convergence of discrete to continuum self-repelling motion[26].The second is whether there are interesting con-nections between the BW and super-Brownian motions;in this regard,the work of[8,9]may be relevant since it deals with noncrossing paths.We continue the Introduction by discussing coalescing random walks and their scaling limits.Let us begin by constructing random paths in the plane, as follows.Consider the two-dimensional lattice of all points(i,j)with i,j integers and i+j even.Let a walker at spatial location i at time j move right or left at unit speed between times j and j+1if the outcome of a fair coin toss is heads(∆i,j=+1)or tails(∆i,j=−1),with the coin tosses independent for different space-time points(i,j).Figure1depicts a simulation of the resulting paths.The path of a walker starting from y0at time s0is the graph of a sim-ple symmetric one-dimensional random walk,Y y0,s0(t).At integer times,Y y0,s0(t)is the solution of the simple stochastic difference equation,Y(j+1)−Y(j)=∆Y(j),j,Y(s0)=y0.(1.1)Fig.1.Coalescing random walks in discrete time;the horizontal coordinate is space and the vertical coordinate is time.THE BROWNIAN WEB3 Furthermore,the paths of distinct walkers starting from different(y0,s0)’s are automatically coalescing—that is,they are independent of each other until they coalesce(i.e.,become identical)upon meeting at some space-time point.If the increments∆i,j remain i.i.d.but take values besides±1(e.g.,±3), then one obtains nonsimple random walks whose paths can cross each other in space-time,although they still coalesce when they land on the same space-time lattice site.Such systems with crossing paths will be discussed in Sec-tion5(see also[20]).After rescaling to spatial steps of sizeδand time steps of sizeδ2,a singlerescaled random walk(say,starting from0at time0)Y(δ)0,0(t)=δY0,0(δ−2t)converges asδ→0to a standard Brownian motion B(t).That is,by theDonsker invariance principle[10],the distribution of Y(δ)0,0on the space ofcontinuous paths converges asδ→0to standard Wiener measure.The invariance principle is also valid for continuous-time random walks, where the move from i to i±1takes an exponentially distributed time. In continuous time,coalescing random walks are at the heart of Harris’graphical representation of the(one-dimensional)voter model[18]and their scaling limits arise naturally in the physical context of(one-dimensional) aging(see,e.g.,[14,15]).Of course,finitely many rescaled coalescing walks in discrete or continuous time(with rescaled space-time starting points) converge in distribution tofinitely many coalescing Brownian motions.In this paper,we present results concerning the convergence in distribution of the complete collection of the rescaled coalescing walks from all the starting points.Our results are in two main parts:1.A new characterization of the limiting object,the standard BW.2.Convergence criteria,which are applied,in this paper,to coalescing ran-dom walks.As a cautionary remark,we point out that the scaling limit motivating our convergence results does not belong to the realm of hydrodynamic limits of particle systems but rather to the realm of invariance principles.A key ingredient of the new characterization and the convergence is the choice of a space on which the BW measure is defined;this is a space whose elements are collections of paths(see Sections2and3).The convergence criteria and application(see,e.g.,Theorems2.2and6.1)are the BW ana-logues of Donsker’s invariance principle.Like Brownian motion itself,we expect that the BW and its variants will be quite ubiquitous as scaling limits,well beyond the context of coalescing random walks(and our suf-ficient conditions for convergence).One situation where this occurs is for two-dimensional“Poisson webs”[11].Another example is in the area of river4FONTES,ISOPI,NEWMAN AND RAVISHANKARbasin modelling;in[24](see also[16,21,27]),coalescing random walks were proposed as a model of a drainage network.Some of the questions about scaling in such models mayfind answers in the context of their scaling lim-its.For more on coalescing random walk and other models for river basins, see[23,30,31].Much of the construction of the BW(but without convergence results)was already done in the groundbreaking work of Arratia[4,5](see also[6,19]) and then in the work of T´o th and Werner[26]who derived many important properties of the BW(see also[25]and[28];in the latter reference,the BW is introduced in relation to black noise).Arratia,T´o th and Werner all recognized that in the limitδ→0there would be(nondeterministic) space-time points(x,t)starting from which there are multiple limit paths and they provided various conventions(e.g.,semicontinuity in x)to avoid such multiplicity.An important feature of our approach to the BW is to accept the intrinsic nonuniqueness by choosing an appropriate metric space in which the BW takes its values.Roughly speaking,instead of using some convention to obtain a process that is a single-valued mapping from each space-time starting point to a single path from that starting point,we allow multivalued mappings;more accurately,our BW value is the collection of all paths from all starting points.This choice of space is very much in the spirit of earlier work[1,2,3]on spatial scaling limits of critical percolation models and spanning trees,but modified for our particular space-time setting;the directed(in time)nature of our paths considerably simplifies the topological setting compared to[1,2,3].The Donsker invariance principle implies that the distribution of any con-tinuous(in the sup-norm metric)functional of Y(δ)0,0converges to that forBrownian motion.The classic example of such a functional is the randomwalk maximum,sup0≤t≤1Y(δ)0,0(t).An analogous example for coalescing ran-dom walks is the maximum over all rescaled walks starting at(or passing through)some vertical(time-like)interval,that is,the maximum value(for times t∈[s,1])over walks touching any space-time point of the form(0,s) for some s∈[0,1].In this case,the functional is not quite continuous for our choice of metric space,but it is continuous almost everywhere(with respect to the BW measure),which is sufficient.The rest of the paper is organized as follows.Section2contains two the-orems.Thefirst,Theorem2.1,is a characterization of the BW,as in[5,26] but adapted to our choice of space;the second and one of our main results is Theorem2.2which is a convergence theorem for the important special case where,even before taking a limit,all paths are noncrossing.Section3con-tains propositions related to Theorem2.1,as well as an alternative charac-terization,Theorem3.1,in which a kind of separability condition is replaced by a minimality condition.In Section4,we present our new characterizationTHE BROWNIAN WEB5 results(Theorems4.1and4.2)based on certain counting random variables, which will be needed for the derivation of our main convergence results.We remark that there are analogous characterization and convergence results jointly for the BW and its dual web of backward paths(important prop-erties of the BW and its dual may be found in[25,26];see also[12]).In Section5,we extend our convergence results to cover the case of crossing paths;the proof of the noncrossing result,Theorem2.2,is given here as a corollary of the more general result.In Section6,we apply our(noncross-ing)convergence results to the case of coalescing random walks.There are two appendices:thefirst covers issues of measurability,the second issues of compactness and tightness.A number of theorems and propositions in this paper are either essentially contained in or easily derived from[5]and/or[26].In those cases,we have omitted the proofs and simply refer the reader to the papers cited above. Detailed proofs can be found in a previous longer version of this paper[12], which uses the same choice of space and notation as this paper.2.Convergence for noncrossing paths.We begin with three metric spaces: (W(or its distributionµ),we define thefinite-dimensional distributions ofW whose distribution is uniquely determined by the following three properties:6FONTES,ISOPI,NEWMAN AND RAVISHANKAR (o)From any deterministic point(x,t)in R2,there is almost surely a unique path W x,t starting from(x,t).(i)For any deterministic n,(x1,t1),...,(x n,t n),the joint distribution ofW x1,t1,...,W xn,t nis that of coalescing Brownian motions(with unit diffusionconstant).(ii)For any deterministic,dense countable subset D of R2,almost surely,W which also touch some point in[a,b]×{t0}.Let alsoˆη(t0,t;a,b)=η(t0,t;a,b)−1.We note that by duality arguments(see[5,26]),it can be shown that for deterministic t0,t,a,b,thisˆηis equidistributed with the number of distinct points in[a,b]×{t0+t}that are touched by paths inW of thestandard BW.(I1)There existθy n∈X n for y∈R2satisfying:for any deterministic y1,...,y m∈D,θy1n,...,θy m n converge in distribution as n→∞to coalescing Brow-nian motions(with unit diffusion constant)starting at y1,...,y m.(B1)∀t>0,lim sup n→∞sup(a,t0)∈R2µn(ˆη(t0,t;a,a+ε)≥1)→0asε→0+.(B2)∀t>0,ε−1lim sup n→∞sup(a,t0)∈R2µn(ˆη(t0,t;a,a+ε)≥2)→0asε→0+.Convergence of coalescing random walks(in discrete and continuous time) (see Theorem6.1)is obtained as a corollary to Theorem2.2.THE BROWNIAN WEB73.Construction and initial characterizations.In this section,we discuss the construction of the BW and the proof of Theorem2.1.Then we give in Theorem3.1a somewhat different characterization of the BW distribution. Let(Ω,F,P)be a probability space where an i.i.d.family of standard Brownian motions(B j)j≥1is defined.Let D={(x j,t j),j≥1}be a count-able dense set in R2.Let W j be a Brownian path starting at position x j at time t j.More precisely,(3.1)W j(t)=x j+B j(t−t j),t≥t j.We now construct,following[5],coalescing Brownian paths out of the family of paths(W j)j≥1by specifying coalescing rules.When two paths meet for thefirst time,they coalesce into a single path,which is that of the Brownian motion with the lower index.We denote the coalescing Brownian paths by W j,j≥1.Notice that the strong Markov property of Brownian motion allows for a lot of freedom in giving a coalescing rule.Any rule,even nonlocal,that does not depend on the realization of the(W j)’s after the time of coalescence will yield the same object in distribution.General definitions and constructions of coalescing Brownian motions can be found in[5].We define the BW skeleton W(D)with starting set D by(3.2)W k=W k(D)={ W j:1≤j≤k},(3.3)W=W(D)= k W k.Now we give detailed definitions of the three spaces introduced in Sec-tion2.(tanh(x2)1+|t1|−R2may be thought of as the image of[−∞,∞]×[−∞,∞]under the mapping (x,t) (Φ(x,t),Ψ(t))≡ tanh(x)R2starting at(f(t0),t0).For(f,t0)in Π,we denote byˆf the function that extends f to all[−∞,∞]by setting it equal to f(t0)for t<t0.Then we take(3.7)d((f1,t1),(f2,t2))= sup t|Φ(ˆf1(t),t)−Φ(ˆf2(t),t)| ∨|Ψ(t1)−Ψ(t2)|.8FONTES,ISOPI,NEWMAN AND RAVISHANKAR(Π,d)is a complete separable metric space.Let now H denote the set of compact subsets of(Π,d),with d H the induced Hausdorffmetric,that is,d H(K1,K2)=supg1∈K1infg2∈K2d(g1,g2)∨supg2∈K2infg1∈K1d(g1,g2).(3.8)(H,d H)is also a complete separable metric space.Definition3.1.W(D)satisfies properties(o)and(i)of Theorem2.1; that is,itsfinite-dimensional distributions(whether from points in D or not) are those of coalescing Brownian motions.The next result is contained in Proposition B.5.Proposition3.2.W(D)=lim k→∞W k(D),where the limit is taken in H.Remark3.2.It can be shown by the methods discussed in Remark B.1 that,almost surely,all paths in2.Proposition3.3.The distribution ofW(D)satisfies property(ii)of Theorem2.1. The next theorem provides an alternative characterization to Theorem 2.1.Other characterizations that will be used for our convergence results, are presented in Section4.Definition3.2(Stochastic ordering).µ1<<µ2if,for g any bounded measurable function on(H,F H)that is increasing[i.e.,g(K)≤g(K′)when K⊆K′], g dµ1≤ g dµ2.Theorem3.1.There is an(H,F H)-valued random variableW≪µW∗.Proof.The proof of this theorem follows easily from Theorem2.1.THE BROWNIAN WEB94.Characterization via counting.In this section,we give other charac-terizations of the BW that will be used for our convergence theorem.Theywill be given in terms of the counting random variablesηandˆηdefined in Definition2.1.We begin with some properties of the BW as constructed in Section1.Proposition4.1.For a BW skeleton W(D),the corresponding count-ing random variableˆηD=ˆηD(t0,t;a,b)satisfies(4.1)P(ˆηD≥k)≤P(ˆηD≥k−1)P(ˆηD≥1)(4.2)≤(P(ˆηD≥1))k=(Θ(b−a,t))k,whereΘ(b−a,t)is the probability that two independent Brownian motions starting at a distance b−a apart at time zero will not have met by time t (which itself can be expressed in terms of a single Brownian motion).Thus,ˆηD is almost surelyfinite and E(ˆηD)<∞.Proof.The proof of this proposition for k=2is contained in[26].The proof for k>2can be readily obtained by using the FKG inequalities—see[12].(The following remark notes that stronger bounds may be obtain-able by the methods of[26].)Remark4.1.By analogy with the number of crossings in the scaling limit of percolation and other statistical mechanics models[2],one may expect the actual decay to be Gaussian in k rather than exponential,asin(4.2).Indeed,as noted by an Associate Editor,probably an upper bound√of the form C k[(b−a)/W(D),there exists a pathθε=(g,t′0)in the skeleton W(D)such that g(s)=f(s)for all s≥t0+ε.The proof of the next proposition follows essentially from Propositions 4.1and4.2.Proposition4.3.Letˆη=ˆη(t0,t;a,b)be the counting random variable for10FONTES,ISOPI,NEWMAN AND RAVISHANKARTheorem4.1.Let W′be an(H,F H)-valued random variable;its dis-tribution equals that of the(standard)BWW.For purposes of proving our convergence results,we will use a modified version of the above characterization theorem in which conditions(o),(i), (ii′′)are all weakened.Theorem4.2.Let W′be an(H,F H)-valued random variable and let D be a countable dense deterministic subset of R2and for each y∈D,let θy∈W′be some single(random)path starting at y.W′is equidistributed with the(standard)BWW,that is,P(ηW′≥k)≤P(ηW.Letη′be the counting random variable appearing in condition(ii′′′)for µ′.Choose some deterministic dense countable subset D and consider the countable collection W∗of paths of W′starting from D.By condition(i′), W∗is equidistributed with our constructed BW skeleton W(based on the same D)and hence the closureW.To complete the proof,we will use condition(ii′′′)to show that W′\W∗already satisfies condition (ii′′′)since W∗were nonempty(with strictly positive probability),then there would have to be some rational t0,t,a,b for whichη′>η∗.But thenP(η′(t0,t;a,b)>η∗(t0,t;a,b))>0(4.5)for some rational t0,t,a,b,and this together with the fact that P(η′≥η∗)= 1(which follows fromW can be replaced by E(ηW′)≤E(ηW )=1+(b−a)/√πt for all t0,t,a,b,then W′is equidis-tributed with the BW.THE BROWNIAN WEB 115.General convergence results.In this section,we state and prove The-orem 5.1,which is an extension of our convergence result for noncrossing paths,Theorem 2.2,to the case where paths can cross (before the scaling limit has been taken).At the end of the section,we show that the noncross-ing Theorem 2.2follows from Theorem 5.1and other results.Before stating our general theorem that allows crossing,we briefly discuss some systems with crossing paths,to which it should be applicable (see also Section 1.3of [5]and [20]).Consider the stochastic difference equation (1.1)where the ∆i,j ’s are i.i.d.integer-valued random variables,with zero mean and finite nonzero variance.Allowing (i,j )to be arbitrary in Z 2,we obtain as a natural generalization of Figure 1a collection of random piecewise linear paths that can cross each other,but that still coalesce when they meet at a lattice point in Z 2.With the natural choice of diffusive space-time scaling and under condi-tions of irreducibility and aperiodicity (to ensure that the walks from any two starting points have a strictly positive probability of coalescing),the scaling limit of such a discrete time system should be the standard BW.To see what happens in reducible cases,consider simple random walks (∆i,j =±1),where the paths on the even and odd subsets of Z 2are independent of each other,and so the scaling limit on all of Z 2consists of the union of two independent BWs.For ∆i,j =±2,the limit would be the union of four independent BWs.We remark that for continuous-time random walks (as discussed in the next section of this paper for ∆i,j =±1),no aperiodicity condition is needed.We proceed with some definitions needed for our general convergence theorem.For a,b,t 0∈R ,a <b ,and t >0,we define two real-valued measur-able functions l t 0,t ([a,b ])and r t 0,t ([a,b ])on (H ,F H )as follows.For K ∈H ,l t 0,t ([a,b ])evaluated at K is defined as inf {x ∈[a,b ]|∃y ∈R and a path in K which touches both (x,t 0)and (y,t 0+t )}and r t 0,t ([a,b ])is defined simi-larly with the inf replaced by sup.We also define the following functions on (H ,F H )whose values are subsets of R .As before,we let K ∈H and suppress K on the left-hand side of the formula for ease of notation:N t 0,t ([a,b ])={y ∈R |∃x ∈[a,b ]and a path in K which (5.1)touches both (x,t 0)and (y,t 0+t )},N −t 0,t ([a,b ])={y ∈R |there is a path in K which (5.2)touches both (l t 0,t ([a,b ]),t 0)and (y,t 0+t )},N +t 0,t([a,b ])={y ∈R |there is a path in K which (5.3)touches both (r t 0,t ([a,b ]),t 0)and (y,t 0+t )}.Remark 5.1.We notice that |N t 0,t ([a,b ])|=η(t 0,t ;a,b ).12FONTES,ISOPI,NEWMAN AND RAVISHANKARLet {X m }be a sequence of (H ,F H )-valued random variables with distri-butions {µm }.We define conditions (B1′),(B2′)as follows.(B1′)∀β>0,lim sup m →∞sup t>βsup t 0,a ∈R µm (|N t 0,t ([a −ε,a +ε])|>1)→0as ε→0+.(B2′)∀β>0,1W .Theorem 5.1is proved through a series of lemmas.Lemma 5.1.Let µbe a subsequential limit of {µm }and suppose that µsatisfies condition (i ′)of Theorem 4.2and:(B1′′)∀β>0,sup t>βsup t 0,a µ(|N t 0,t ([a −ε,a +ε])|>1)→0as ε→0+,(B2′′)∀β>0,1W .We need to show that it also satisfies condition (ii ′′′)of Theorem 4.2.Let a <b,t 0∈R and t >0be given.For the random variable X we will denote the counting randomvariable η(t 0,t ;a,b )by ηand the corresponding variable for X ′by η′.Let z j =(a +j (b −a )/M,t 0),for j =0,1,...,M ,be M +1equally spaced points in the interval [a,b ]×{t 0}.THE BROWNIAN WEB13 Now defineηM=|{x∈R|∃a path in X which touches both a point in {z0,...,z M}and(x,t+t0)}|,where|·|stands for cardinality.Letη′M be thecorresponding random variable for X′.Clearly,η≥ηM andη′≥η′M.From (B1′′)it follows thatηM=η′M almost surely.Now letε=(b−a)8,0≤δ<tεlim supε′→0sup t>βsup t0,alim supδ→0µ(O(a,t0,t,ε,ε′,δ))→0asε→0+.Proof.We prove the lemma by showing that conditions(i′)and(B1′′) together with(B2′′′)imply condition(B2′′).Letβ>0.Define C1(b,t0,ε′,δ) as{K∈H|there is a path in K which touches both(b,t0)and{b−ε′}×[t0,t0+δ]∪{b+ε′}×[t0,t0+δ]},and C2(a,t0,ε,ε′,δ)as{K∈H|there is a path in K which touches both[a−ε,a+ε]×{t0} and{a−ε−ε′}×[t0,t0+δ]∪{a+ε+ε′}×[t0,t0+δ]}. Now observe that(modulo sets of zeroµmeasure){N t0,t([a−ε,a+ε])=N+t0,t([a−ε,a+ε])∪N−t0,t([a−ε,a+ε])}∩C c1(a+ε,t0,ε′,δ)∩C c1(a−ε,t0,ε′,δ)∩C c2(a,t0,ε,ε′,δ)∩{|N t0+δ,t−δ([a−ε−2ε′,a−ε+2ε′])|=1}∩{|N t0+δ,t−δ([a+ε−2ε′,a+ε+2ε′])|=1}⊆O(a,t0,t,ε,ε′,δ).14FONTES,ISOPI,NEWMAN AND RAVISHANKAR Therefore,we haveµ(N t0,t ([a−ε,a+ε])=N+t0,t([a−ε,a+ε])∪N−t0,t([a−ε,a+ε]))≤µ(O(a,t0,t,ε,ε′,δ))+µ(C2(a,t0,ε,ε′,δ))+µ(C1(a+ε,t0,ε′,δ))+µ(C1(a−ε,t0,ε′,δ))+µ(|N t0+δ,t−δ([a−ε−2ε′,a−ε+2ε′])|>1)+µ(|N t0+δ,t−δ([a+ε−2ε′,a+ε+2ε′])|>1).Lettingδ→0,we obtainµ(N t0,t ([a−ε,a+ε])=N+t0,t([a−ε,a+ε])∪N−t0,t([a−ε,a+ε]))≤lim supδ→0{µ(O(a,t0,t,ε,ε′,δ))+µ(C2(a,t0,ε,ε′,δ))+µ(C1(a+ε,t0,ε′,δ))+µ(C1(a−ε,t0,ε′,δ)) (5.4)+µ(|N t0+δ,t−δ([a−ε−2ε′,a−ε+2ε′])|>1)+µ(|N t0+δ,t−δ([a+ε−2ε′,a+ε+2ε′])|>1)}. Now,lim δ→0µ(C1(a+ε,t0,ε′,δ))=limδ→0µ(C1(a−ε,t0,ε′,δ))=limδ→0µ(C2(a,t0,ε,ε′,δ))=0,since elements of H are compact subsets K ofΠ,and compact sets of con-tinuous functions are equicontinuous.If the above limit did not vanish,then there would be positiveµ-measure for K to contain paths with arbitrarily close toflat segments,thus violating equicontinuity.Now since,t−δ>t2,it follows from(B1′′)thatsup t>βsupa,t0sup0<δ<t/2µ(|N t0+δ,t−δ([a−γ,a+γ])|>1)≤supt>β/2supa,t0µ(|N t0,t([a−γ,a+γ])|>1)→0asγ→0.This implies that for allε>0,lim sup ε′→0supt>βsupa,t0lim supδ→0µ(|N t0+δ,t−δ([a±ε−2ε′,a±ε+2ε′])|>1)=0.(5.5)Together with(5.4),this gives ussup t>βsupt0,aµ(N t0,t([a−ε,a+ε])=N+t0,t([a−ε,a+ε])∪N−t0,t([a−ε,a+ε]))(5.6)≤lim supε′→0supt>βsupa,t0lim supδ→0µ(O(a,t0,t,ε,ε′,δ)).THE BROWNIAN WEB15 Now,using(B2′′′),we obtain1W .From Lemmas5.1and5.2it follows thatit is sufficient to prove condition(i′)of Theorem4.2,condition(B1′′)and condition(B2′′′).Letβ>0and define for all0≤δ<t16FONTES,ISOPI,NEWMAN AND RAVISHANKAR It follows from(B1′)thatlim supmsupt>β/2supt0,aµm(|N t0,t([a−2ε,a+2ε])|>1)→0asε→0+.This proves(B1′′),which implies that:(o)starting from any deterministic point,there isµ-almost surely only a single path in X.Combining this with(I1),we readily obtain that:(i)thefinite-dimensional distributions of X are those of coalescing Brow-nian motions with unit diffusion constant.Condition(i′)of Theorem 4.2follows immediately from(o)and(i).Now we proceed to verify condition(B2′′′).We havesup t>βsupt0,alim supδ→0µ(O(a,t0,t,ε,ε′,δ))≤supt>β/2supa,t0µ(O(a,t0,t,ε,ε′,0))≤lim supmsupt>β/2supa,t0µm(O(a,t0,t,ε,ε′,0))≤lim supmsupt>β/2supa,t0µm(N t0,t([a−ε−ε′,a+ε+ε′])=N+t0,t([a−ε−ε′,a+ε+ε′])∪N−t0,t([a−ε−ε′,a+ε+ε′])),where the second inequality follows from the fact that O(a,t0,t,ε,ε′,0)is an open subset of H.For the third inequality to hold we need to ensure that there is no more than one path touching either(a−ε−ε′,t0)or(a+ε+ε′,t0);this follows from(B1′).Sinceε′<εεlim supε′→0supt0,alim supδ→0µ(O(a,t0,t,ε,ε′,δ))→0asε→0+,proving condition(B2′′′).This completes the proof of the theorem.We now suppose that X1,X2,...is a sequence of(H,F H)-valued random variables so that each X i consists of noncrossing paths.The noncrossing condition produces a considerable simplification of Theorem5.1,namely, Theorem2.2.Proof of Theorem2.2.This is an immediate consequence of Re-mark5.2,Theorem5.1and Proposition B.2.THE BROWNIAN WEB 176.Convergence for coalescing random walks.We now apply Theorem 2.2to coalescing random walks.For that,we begin by precisely defining Y (resp. Y),the set of all discrete-(resp.continuous-)time coalescing random walks on Z .For δan arbitrary positive real number,we obtain sets of rescaled walks,Y (δ)and Y(δ),by the usual rescaling of space by δand time by δ2.The (main)paths of Y are the discrete-time random walks Y y 0,s 0,as described in the Introduction and shown in Figure 1,with (y 0,s 0)=(i 0,j 0)∈Z ×Z ar-bitrary except that i 0+j 0must be even.Each random walk path goes from (i,j )to (i ±1,j +1)linearly.In addition to these,we add some boundary paths so that Y will be a compact subset of Π.These are all the paths of the form (f,s 0)with s 0∈Z ∪{−∞,∞}and f ≡∞or f ≡−∞.Note that for s 0=−∞there are two different paths starting from the single point at s 0=−∞int )k −1,whichyields (B1)and (B2)as desired.APPENDIX A:Some measurability issues.Let (H ,d H )denote the Hausdorffmetric space induced by (Π,d ).F H denotes the σ-field generated by the open sets18FONTES,ISOPI,NEWMAN AND RAVISHANKARof H.We will consider now cylinders of H.Let usfix nonempty horizontal segments I1,...,I n in R2(i.e.,I k=I′k×{t k}),where each I′k is an interval (which need not befinite and can be open,closed or neither)and t k∈R. DefineC t0I1,...,I n :={K∈H:there exists(f,t)∈K such that(A.1)t>t0and(f,t)goes through I1,...,I n},THE BROWNIAN WEB19 with r−≤r+.For sfixed,let x−(s)=x−(s,ε)and x+(s)=x+(s,ε)be the solutions for smallεoftanh(x)−tanh(ˆf(s))C s I,where I={I1,...,I m};Iℓ=[x−i(s′ℓ),x+i(s′ℓ)]×{s′ℓ}; s≤s′1≤s′2≤···;s,s′1∈[r−i,r+i];and each s′ℓ∈S.Note that in the definition (A.2)of such aB H(K0,ε).An explanation is presented immediately after the formula:。

Book Club DiscussionsBook Club Discussions Book clubs have become increasingly popular in recent years as a way for individuals to come together and discuss their love for literature. These gatherings provide a platform for readers to engage in meaningful conversations, share their thoughts and opinions, and gain new perspectives on the books they read. In this article, we will explore the benefits of book club discussions from multiple perspectives. From a personal standpoint, participating in a book club discussion can be a highly enriching experience. It allows individuals to delve deeper into the themes, characters, and plotlines of a book, gaining a greater understanding and appreciation for the author's work. The opportunity to hear different viewpoints and interpretations can broaden one's own perspective, fostering critical thinking and analytical skills. Moreover, bookclub discussions provide a sense of community and camaraderie among members, creating lasting friendships based on a shared passion for reading. On a broader level, book club discussions contribute to the promotion of literacy and the cultivation of a reading culture. By engaging in thoughtful conversations about books, participants develop their reading skills and expand their literary knowledge. This, in turn, encourages them to explore a wider range of genres and authors, fostering a lifelong love for reading. Additionally, book clubs often introduce members to books they may not have otherwise chosen, exposing them to diverse voices and narratives. Book club discussions also offer a platform for authors to connect with their readers on a more personal level. Authors may choose to participate in book club meetings or engage with readers through online forums, allowing them to gain valuable insights into their readers' thoughts and reactions. This direct interaction can be immensely rewarding for both authors and readers,as it provides an opportunity for deeper engagement and mutual appreciation. From an educational perspective, book club discussions can be a valuable tool in classrooms. They encourage students to actively engage with the text, fostering critical thinking and analytical skills. By participating in discussions, students learn to articulate their thoughts and opinions, listen to others respectfully,and constructively challenge ideas. These skills are transferable to other areasof their academic and personal lives, making book club discussions a valuableaddition to any educational curriculum. In conclusion, book club discussionsoffer numerous benefits from personal, societal, author-reader, and educational perspectives. They provide a platform for individuals to connect, share ideas, and gain new insights into the books they read. Whether it's fostering a love for reading, promoting literacy, or facilitating meaningful interactions between authors and readers, book club discussions play a vital role in enriching the literary landscape. So, why not join a book club and embark on a journey of exploration and discovery through the power of collective reading and discussion?。

行为经济学教材
以下是一些常用的行为经济学教材：
1.《行为经济学》（Behavioral Economics）- 埃尔南·阿兰奇（Hernán Arance）
2.《行为经济学》（Behavioral Economics）- 马克·海尔德雷斯（Mark Hadenrees）
3.《行为经济学》（Behavioral Economics）- 巴勃罗·达席尔瓦·博尔哈斯（Pablo Dias Berdichevsky）
4.《行为经济学》（Behavioral Economics）- 马修·鲍尔斯（Matthew Rabin）
5.《行为经济学导论》（Introduction to Behavioral Economics）- 尤斯·雅格拉斯（José Ignacio Cuesta）
6.《行为经济学与计算机科学》（Behavioral Economics and Computer Science）- 理查德·斯通（Richard Stone）
7.《行为金融学》（Behavioral Finance） - 赫博伊·谢夫兰（Hersh Sheffrin）
这些教材覆盖了行为经济学的基本原理和理论，包括心理学，经济学和行为科学的交叉领域。

读者可以选择适合自己的水平和兴趣的教材进行学习。

班博思市场营销培训教程(英文版)Barbosa Market Marketing Training TutorialIntroduction:Welcome to the Barbosa Market Marketing Training Tutorial! In this tutorial, we will take you through the fundamental concepts of marketing and provide you with the essential skills and knowledge to excel in the field of market marketing. Whether you are a beginner or an experienced professional, this tutorial will equip you with the tools to develop effective marketing strategies and achieve your business goals.Chapter 1: Understanding Market Marketing1.1 Definition and Importance of Marketing:- Definition: What is market marketing and why is it crucial for businesses?- Evolution of marketing: From traditional to digital marketing. 1.2 Elements of Market Marketing:- Product: Understanding the importance of product development and differentiation.- Price: Determining the right pricing strategy based on market research.- Place: Choosing the right distribution channels to reach the target audience.- Promotion: Effective utilization of advertising, public relations, and sales promotion to enhance brand visibility.Chapter 2: Market Research2.1 Importance of Market Research:- Understanding customer needs, preferences, and behaviors.- Identifying market trends and competition analysis.2.2 Conducting Market Research:- Gathering primary and secondary research data.- Analyzing and interpreting research findings.Chapter 3: Target Audience Identification3.1 Segmentation:- Dividing the market into segments based on demographics, psychographics, and behavior.- Identifying niche markets for effective targeting.3.2 Targeting:- Selecting the most profitable segments to focus on.- Creating buyer personas to better understand the target audience. Chapter 4: Developing Marketing Strategies4.1 Creating Competitive Advantage:- Identifying unique selling propositions (USPs).- Conducting a SWOT analysis to leverage strengths and overcome weaknesses.4.2 Setting Marketing Objectives:- Defining SMART (Specific, Measurable, Achievable, Relevant,Time-bound) objectives.- Aligning marketing objectives with business goals.4.3 Designing Marketing Mix:- Product: Features, benefits, and branding strategies.- Price: Pricing strategies and value propositions.- Place: Determining the right distribution channels.- Promotion: Developing integrated marketing communication plans.Chapter 5: Implementing Marketing Strategies5.1 Executing the Marketing Plan:- Allocating resources and developing a marketing budget.- Implementing marketing activities across various channels.5.2 Monitoring and Evaluation:- Tracking key performance indicators (KPIs) to measure marketing effectiveness.- Making necessary adjustments based on evaluation results. Chapter 6: Digital Marketing6.1 Introduction to Digital Marketing:- Overview of digital marketing channels and tools.- The role of social media, search engine optimization (SEO), and content marketing.6.2 Creating Digital Marketing Campaigns:- Designing engaging website content and optimizing userexperience.- Harnessing the power of email marketing and online advertising. Conclusion:Congratulations on completing the Barbosa Market Marketing Training Tutorial! We hope this comprehensive guide has provided you with valuable insights into the world of market marketing and equipped you with the necessary skills to thrive in this field. Remember, effective marketing strategies are essential in driving business growth and achieving success. Keep learning, exploring, and adapting to stay ahead in the ever-evolving world of marketing.Chapter 7: Marketing Communication7.1 The Role of Marketing Communication:- Understanding the importance of effective communication in marketing.- Building brand awareness and enhancing brand reputation through communication.7.2 Integrated Marketing Communication:- Creating a consistent message across different communication channels.- Utilizing various communication tools such as advertising, public relations, sales promotion, and personal selling.7.3 Developing a Communication Strategy:- Identifying the target audience and understanding their communication preferences.- Selecting the most appropriate communication channels forreaching the target audience.7.4 Crafting a Compelling Message:- Creating attention-grabbing headlines and compelling content. - Using storytelling techniques to engage the audience and evoke emotions.Chapter 8: Marketing Analytics8.1 Importance of Marketing Analytics:- Utilizing data-driven insights to make informed marketing decisions.- Measuring the effectiveness of marketing campaigns and optimizing strategies.8.2 Key Marketing Metrics:- Understanding key performance indicators (KPIs) such as ROI, conversion rate, customer acquisition cost, and customer lifetime value.- Setting benchmarks and targets for measuring marketing performance.8.3 Analyzing Marketing Data:- Gathering and organizing marketing data from various sources. - Utilizing analytics tools to analyze data and extract meaningful insights.Chapter 9: Customer Relationship Management9.1 Building Strong Customer Relationships:- Understanding the importance of customer satisfaction and loyalty.- Implementing strategies to enhance customer experience and build long-term relationships.9.2 Customer Retention:- Developing loyalty programs and incentives to encourage repeat purchases.- Implementing effective customer support systems to address customer concerns.9.3 Customer Feedback and Insights:- Gathering customer feedback through surveys and reviews.- Utilizing customer insights to improve products, services, and marketing strategies.Chapter 10: Ethics and Social Responsibility in Marketing10.1 Ethical Marketing Practices:- Adhering to ethical standards in marketing communications and advertising.- Avoiding deceptive practices and ensuring transparency in marketing strategies.10.2 Socially Responsible Marketing:- Incorporating social and environmental considerations into marketing practices.- Supporting social causes and community initiatives to build a positive brand image.10.3 Balancing Profit with Purpose:- Achieving business objectives while maintaining ethical and socially responsible practices.- Establishing a reputation as a responsible and trustworthy brand. Conclusion:In this tutorial, we have covered key concepts and strategies in market marketing. From understanding the definition and importance of marketing to developing effective marketing strategies and leveraging digital marketing tools, this tutorial has equipped you with the essential knowledge and skills to excel in the field of market marketing. Additionally, we have explored topics such as market research, target audience identification, marketing communication, marketing analytics, customer relationship management, and ethics and social responsibility in marketing. By applying these concepts and strategies, you will be able to develop and implement successful marketing campaigns that drive business growth and achieve your objectives. Keep learning, adapting, and staying updated with the latest marketing trends to remain competitive in today's dynamic business landscape. Best of luck in your marketing journey!。

2015版恩波38套英语的解析质量全文共3篇示例，供读者参考篇1Analysis of the Quality of the Enbo 38 Sets of English in 2015IntroductionThe Enbo 38 Sets of English, published in 2015, is a popular series of English textbooks used by students in China. These textbooks cover a wide range of topics and are designed to help students improve their English language skills. In this analysis, we will evaluate the quality of the Enbo 38 Sets of English based on various criteria such as content, language proficiency, and effectiveness in teaching English to students.ContentOne of the key aspects of the Enbo 38 Sets of English is its content. The textbooks cover a variety of topics ranging from everyday conversations to more complex grammar structures. Each set contains a mix of reading passages, listening exercises, vocabulary lists, grammar explanations, and writing tasks. The content is well-organized and progresses in difficulty from set toset, allowing students to gradually build their English language skills.Language ProficiencyThe Enbo 38 Sets of English aims to help students improve their language proficiency in all four skills – speaking, listening, reading, and writing. The textbooks provide ample opportunities for students to practice their speaking and listening skills through dialogues, audio recordings, and role-playing activities. Reading passages are diverse and cover a range of genres and topics, helping students improve their reading comprehension. Writing tasks are also included in each set, allowing students to practice and improve their writing skills.Effectiveness in TeachingThe effectiveness of the Enbo 38 Sets of English in teaching English to students is quite high. The textbooks are designed to be engaging and interactive, with colorful illustrations, real-life scenarios, and practical exercises that help students apply their language skills in various contexts. The progression in difficulty from set to set ensures that students continuously challenge themselves and improve their English skills over time. Additionally, the textbooks provide teachers with a variety of resources and materials to supplement their lessons, making iteasier for them to teach and reinforce key language concepts to students.ConclusionOverall, the Enbo 38 Sets of English is a high-quality series of textbooks that effectively help students improve their English language skills. The content is comprehensive andwell-organized, the language proficiency activities are diverse and engaging, and the effectiveness of teaching English to students is evident. For students in China looking to improve their English proficiency, the Enbo 38 Sets of English is a highly recommended resource.篇2Analyzing the Quality of the 2015 Edition of Enbo 38 Sets of EnglishIntroductionEnbo 38 Sets of English is a widely used textbook in China for teaching English as a second language. The 2015 edition has been considered by many students, teachers, and academics as an important tool for learning English. In this analysis, we will examine the quality of the 2015 edition of Enbo 38 Sets of English.ContentThe content of the 2015 edition of Enbo 38 Sets of English is comprehensive and covers various aspects of English language learning, including grammar, vocabulary, reading, writing, listening, and speaking. Each set includes different exercises and activities that help students practice and improve their language skills. The topics are relevant and engaging, making learning English enjoyable for students.StructureThe structure of the textbook is well-organized, with each set focusing on a specific topic or language skill. The progression of difficulty level is logical and allows students to develop their English proficiency gradually. The instructions provided are clear and easy to follow, making it easier for students to learn independently or in a classroom setting.AuthenticityThe 2015 edition of Enbo 38 Sets of English includes authentic materials such as articles, dialogues, and audio recordings that reflect real-life situations and contexts. This helps students to understand and use English in everyday communication. The use of authentic materials also exposesstudents to different accents, dialects, and cultural aspects of English-speaking countries.AssessmentThe textbook includes a variety of assessment tasks, such as quizzes, tests, and exercises, that help students evaluate their progress and identify areas for improvement. The assessment tasks are well-designed and align with the learning objectives of each set. This allows teachers to monitor students' performance and provide timely feedback to support their learning.Technology IntegrationThe 2015 edition of Enbo 38 Sets of English integrates technology by providing access to online resources, interactive activities, and multimedia materials that enhance students' learning experience. Students can practice listening and speaking skills, review grammar and vocabulary, and engage with English language content through digital platforms.ConclusionOverall, the 2015 edition of Enbo 38 Sets of English is a high-quality textbook that effectively supports English language learning for students in China. The comprehensive content, well-organized structure, authenticity of materials, assessmenttasks, and technology integration make it a valuable resource for teachers and students. This textbook has contributed to the improvement of English language education in China and has helped students develop their language skills effectively.篇3Title: Analysis of the Quality of Enbo 38 sets of English in 2015Introduction:Enbo 38 sets of English textbooks have been widely used in primary and middle schools in China since 2015. These textbooks are known for their comprehensive coverage of various English topics and exercises. In this document, we will analyze the quality of Enbo 38 sets of English in 2015.Content:1. Content Analysis:The content of Enbo 38 sets of English is designed to meet the requirements of the English curriculum in China. The textbooks cover a wide range of topics including vocabulary, grammar, reading, writing, listening, and speaking. The contentis presented in a systematic and organized manner, making it easier for students to understand and learn.2. Exercises and Activities:Enbo 38 sets of English include a variety of exercises and activities to help students practice and apply what they have learned. The exercises are designed to be engaging and interactive, making learning fun and enjoyable for students. The activities also encourage students to think critically and creatively, enhancing their language skills.3. Language Skills Development:The Enbo 38 sets of English are designed to help students develop their language skills in all aspects. The textbooks focus on building students' vocabulary, improving their grammar, enhancing their reading and writing skills, and developing their listening and speaking abilities. The exercises and activities in the textbooks are specially designed to strengthen these language skills.4. Cultural Awareness:Enbo 38 sets of English also focus on promoting cultural awareness and understanding among students. The textbooks include stories, articles, and activities that introduce students todifferent cultures and traditions from around the world. This helps students develop a broader perspective and appreciate the diversity of the world.5. Teacher and Student Resources:Enbo 38 sets of English provide teachers with comprehensive resources to support their teaching. The textbooks come with teacher's guides, answer keys, and supplementary materials to help teachers effectively deliver the lessons. Students also have access to online resources and interactive activities to further enhance their learning experience.Conclusion:Overall, the quality of Enbo 38 sets of English in 2015 is commendable. The textbooks are well-structured, engaging, and comprehensive, making them an effective tool for teaching and learning English in Chinese schools. With their focus on language skills development, cultural awareness, and teacher and student resources, Enbo 38 sets of English have proven to be a valuable resource for English language education in China.。

2015考研英语阅读理解精读P17—法学类Passage 17“This is not the type of place where this happens," city council president George Carlton told a reporter, after the horror became public in his hometown, Sylacauga, Ala. He echoed what was said in Jasper, Texas, a year ago. Few people then had ever heard of Jasper. A week ago, even fewer could have pointed out Sylacauga on a map. A tiny city of 13,000, halfway between Birmingham and Montgomery, Sylacauga was known for its white marble quarries, textile mills and ice-cream factory. But last week Sylacauga, like Jasper, became a chapter in the recent history of hatred.According to police, Steven Eric Mullins, 25, and Charles Monroe Butler Jr., 21, plotted for two weeks to murder Billy Jack Gaither, 39. On Feb. 19, they arranged to meet him at a Sylacauga bar and lured him to a secluded area. There they beat him and dumped him into the trunk of his car. They then drove about 15 miles to Peckerwood Creek in Coosa County. There, says Coosa County Sheriff's Deputy Al Bradley, "they took him out of the trunk, took an ax handle and beat him to death." They set two old tires aflame, says Bradley, "then they put the body on the fire." They did it all, the deputy says, because Gaither was gay.Gaither's death has become a rallying point for gay-rights organizations' and state legislators' pushing a bill that would extend Alabama's three-year-old hate-crimes law beyond race, color, religion and national origin to cover crimes related to sexual orientation as well. "It's unfortunate that somebody had to lose his life in order for this legislation to pick up momentum here in the state of Alabama," says state Representative Alvin Holmes, who failed to get the original law amended when it was passed in 1996. Holmes filed for extending the law after Matthew Shepard, a gay student, was beaten and left to die on a fence in Wyoming last October, an incident that sparked national outrage. Even Wyoming failed to pass hate-crime legislation in the wake of the Shepard lynching. Like Shepard, Gaither did not hesitate to admit being gay, though he adhered quietly to Sylacauga's Southern dispositions. And friends dispute Mullins' and Butler's allegations that a sexual proposition incited the murder. Gaither's brother Randy told CNN: "Regardless of his personal life or anything, he doesn't deserve to be killed for this.""The message people are getting is that gay people are second-class citizens," says Tracey Conaty, spokesperson for the National Gay and Lesbian Task Force.Before Gaither's murder, activists were planning a major national pro-gay offensive. From March 21 to March 27, the task force will launch its "Equality Begins at Home" campaign, with 250 grass-roots events in all 50 states aimed at passing anti-gay-bashing legislation. Says Conaty: "These laws reflect the conscience of a community and send an important message." The March events, says Urvashi Vaid, director of the task force's policy institute, will involve straight people concerned about neighbors denied basic human rights. Adds Vaid: "It's more than just a gaything."注(1)：本文选自By Sylvester Monroe Time; 03/15/99, Vol. 153 Issue 10, p47, 2/3p, 3c, 1bw 注(2)：本文习题命题模仿对象2003年真题Text 41. What is implied in the first two paragraphs?[A] there are many murders in the recent history of hatred[B]the murder also happened in Jasper one year ago[C] it is another case of the gay being tortured to death[D]the city council president comes from Sylacauga2. The author uses the example of Matthew Shepard to show that ________.[A] it is difficult to extend the hate-crime legislation[B]people want to extend the hate-crime law[C]the gays are really in a terrible fix[D] people are indifferent to the gay student3. Alvin Holmes’attitude toward the gay victims is _________.[A]indifferent[B]sympathetic[C]outrageous[D]considerate4. Similar to Matthew Shepard, Gaither’s death ________.[A]aroused people’s sympathy for the gay[B] sharpened people’s awareness[C]gave legislation some momentum[D]failed to have any change in the legislation5. The text intends to express the idea that __________.[A] people should be concerned about their gay neighbors[B]the gay people shouldn’t be regarded as second-class citizens[C] the legislation for the gay still has a long way to go[D]more pro-gay campaigns should be launched答案：CABDC篇章剖析本文采用提出问题-----分析问题的模式。