BBC HorizonThe Age of Big DataWord文件下载.docx
《BBC HorizonThe Age of Big DataWord文件下载.docx》由会员分享,可在线阅读,更多相关《BBC HorizonThe Age of Big DataWord文件下载.docx(21页珍藏版)》请在冰豆网上搜索。
6am,LosAngeles.ThestartofshiftintheFoothilldivision.
OfficerSteveNunes,a12-year-veteranoftheLAPD,andhispartnerDannyFraserheadouttopatrol.
Rightnow,we'
renorthofLosAngeles,downtownLosAngeles,intheSanFernandoValleyarea.
TheirbeatisoneofLA'
stoughestneighbourhoods.
There'
salotofBFMVs,burglaryfrommotorvehicles.
salotofrobberies,there'
salotofgangandnarcoticactivityoverhere.
salotofpeoplesellingdrugs.
Thegangthat'
sinthisareaarecalledtheProjectBoys.
They'
reaHispanicgang.
Despitetheirexperienceandintimateknowledgeoftheneighbourhood,today,theirpatrolisbeingcontrolledbyacomputeralgorithm.
Youknow,Iwasn'
treallytoohappyaboutit,youknow,speciallyasapoliceofficeryouknow,wekindofgooffofwhatweknowfromourtraining.
Weweren'
ttoohappyaboutacomputertellinguswhereweneedtodoourpoliceworkandwhatareaweneedtodrivearound.
SteveandDannyarepartofaground-breakingtrial.
Anequationisbeingusedtopredictwherecrimewilloccurontheirwatch.
Isawsomepeoplehangingoutbythelaundry,likethelittlelaundry.
Icanseethecorner.
Allright,let'
sstop.
Ifitspredictionsarecorrect,thesystemwillberolledoutacrossallLA.
-Hey,stop,yeah,stop.-Stop.
Putyourhandsonyourhead.
AndthecomputeralgorithmwillbecomearoutinepartofSteve'
sworkinglife.
Spreadyourfeet,faceforward.
-Haveanythingonyou?
-Stopmoving.
YouaProjectBoytooorno?
TheambitiontopredictcrimewasbornoutofaremarkablecollaborationbetweentheLAPD...
andtheUniversityofCalifornia.
JeffBrantinghammightseemanunlikelycrimefighter.
Aprofessorofanthropology,heisanexpertonremotehunter-gatherertribesinChina,buthe'
sconvinced
thatfromremoteChinatoganglandLA,allhumanbehaviourisfarmorepredictable
thanyoumightliketobelieve.
Weallliketothinkthatweareincontrolofeverything,butinfactallofourbehaviourisveryregular,verypatternedinwaysthatisoftenfrighteningtous.
Offendersarenodifferent.
Theydoexactlythesamethingsoverandoverandoveragain,andtheircriminaloffendingpatternsemergerightoutofthatregularityoftheirbehaviour.
JeffbelievedhecouldfindrepeatingpatternsofcriminalbehaviourintheLAPD'
svastdataset-
13millioncrimesrecordedover80years.
TheLAPDhavedrovesanddrovesofdataaboutwhereandwhencrimeshavebeenoccurring.
Itrepresentsatreasuretroveofpotentialinformationforunderstandingthenatureofcrime.
TheLAPDalreadyusetheircrimedatatoidentifyhotspotsofcrime,butthatonlytellsthemwherecrimehasalreadystruck.
We'
vegottenverygoodatlookingatdotsonamap,andwhere,wherecrimehasoccurredandtheproblemwiththatisthat,sometimes,you'
remakinganassumptionthattodayisthesameasyesterday.
JeffBrantinghamplannedtodosomethingmoreradicalandmoreusefulpredictthefuture.
Hebelievedhecouldusepatternsinthecrimedatatopredictwhereandwhencrimewaslikelytooccur.
velongusedthepatternsinnaturetomakepredictions.
Fromthesettingsun,
welearnedwhentoexpectthenewday.
Thephasesofthemoonallowedustoforecasttheebbandflowofthetides.
Andfromobservingthepatternsofthestars,wemasteredtheartofnavigation.
ButJeffBrantinghamwantedtodosomethingfarmoreambitious.
Hewantedtoteaseoutpatternsintheapparentchaosofhumanbehaviour,touncoverthemintheLAPD'
svastdatasetof13millionpastcrimes.
Youcanhavegutfeelingsaboutthecrime,butultimately,youneedtothinkaboutworkinginamathematicalframework.
Becausemathematicsgivesyoutheabilitytounderstandexactlywhythingsarehappeningwithinthedatainawaythatgutfeelingsdonot.
Jeffneededanexpertinpatterndetection.
Heturnedtohiscolleague,UCLAmathematicianGeorgeMohler.
Asmathematicians,we'
reinterestedinunderstandingwhat'
saroundyouso,youknow,howdowavespropagateifyouthrowapebbleintothewater?
Thedistributionoftreesinaforest.
Somathematicalmodels
canhelpyouunderstandthosetypesofthings.
Georgecouldusemathematicaltoolstoseewhatwashiddeninthecrimedata.
Andtherewerehintsofapatterninit.
Whatyouseeisthatafteracrimeoccurs,there'
sanelevatedriskandthatrisktravelstoneighbouringregions.
Sowhatwewantedtodoisdevelopamodeltotakethatintoaccountsopolicecouldmaybeusethatinformationtopreventthosecrimesfromoccurring.
Hestartedwithamathematicalmodelthatwasalreadybeingused,righthereonthewestcoastofAmerica.
SouthernCaliforniaisearthquakecountry.
SittingontheSanAndreasFault,there'
sanaverageof10,000earthquakesandafter-shockseveryyear.
Thebiggestfor100yearswastheLomaPrietaearthquakeof1989.
Itsepicentrewashere,justoutsideSantaCruz,California.
Thereisquitesimplynomathematicalmodelthatcanpredictanearthquakelikethisone.
Butaftertheearthquakecometheafter-shocksandthat'
sadifferentmatter.
Sowe'
reseveralhundredmetresfromtheepicentre.
Nearbywasoneoftheafter-shocksoftheoriginalLomaPrietaearthquake.
Afteralargeearthquakeoccurs,thereisaprobabilitythatanotherearthquakewillfollownearbyinspaceandtime.
Georgediscoveredseismologistshadfoundapatterntoearthquakeafter-shocksanddevelopedanalgorithmtopredicttheseafter-shockclusters.
Thesetypesofclusteringpatternsarealsoseenincrimedata.
So,afteracrimeoccurs,youwillseeanincreasedlikelihoodoffutureeventsnearbyinspaceandtime.
Youcanthinkofthemasafter-shocksofcrime.
GeorgeandJefftooktheequationforpredictingearthquakeafter-shocksandbegantoadaptittopredictcrime.
Sothemodelisbrokenintoseveralparts,sotheoverallrateofcrime,whichwe'
llcallLamda,modelstherateofeventsinspaceandtime.
WeusetheGreekletterMyutorepresentthebackgroundamountofcrimethat'
sgoingon.
ThesecondcomponenttoLamdaisG.
Gmodelsthedistributionofcrimesfollowinganinitialevent.
Thiswholetermoveralldescribeswhatwecallself-excitation,thatacrimethatoccurstodayactuallyself-excitesthepossibilityoffuturecrimes.
SoLamdaequalsMyuplusG,isthatright?
Well,sortof,soLamdaequalsMyuplusGpositionedatallthepasteventsinyourdataset.
GeorgeandJefftooktheiralgorithmbacktothestreetsofLA.
Whentheypluggedtheoldcrimedataintotheequation,itgeneratedpredictionsthatfittedwhathadhappenedinthepast.
Butcoulditalsopredictthefuture?
Theybegantoproducedailycrimeforecasts,identifyinghotspotswherecrimewaslikelytostrikeinthefuture.
-Adam11,Nunes.-Sir
-Thern.-Sir.
-Adam23,Fowler.-Sir.
-Wallier.-Sir.
Let'
sgotothemissionmapsifyouwould,please.
Today,theLAPDisputtingthesepredictionstothetest.
ThecopsinFoothillareassignedboxesofjust500squarefeetwherethealgorithmpredictscrimeismostlikelytooccurintheir12-hourwatch.
Right,predictivemissionfortodayis,we'
vegot15boxesheretoaddress,inAdam11'
sarea,12260FoothillBoulevard.
reinstructedtohittheirboxesasoftenastheycan.
OsborneandFoothillBoulevard.
Soyou'
vegotyourmissionfortheday?
Solet'
sgooutthere,havefunandbesafe.
Yeah,thereisahomicideblinkingupthere.
ThetrialismonitoredattheRealTimeCrimeCentreindowntownLA.
Whatwe'
relookingathereistheforecastthatwasproducedbythePredPolsoftware.
Soifyouseeonthecentreofthismap,we'
vegotthreenearlycontiguousforecastboxesaroundthisarea,andthenanadjacentone.
Sothisisgoodinformationfortheofficers.
Theycangooutthere,workupanddownthatstreet,Sheldon,andsomeofthosesidestreets,andlookforcriminalactivityorevidencethatcriminalactivitymightbeafoot.
OK,Roger,we'
lltakeit.
SteveandDannyhavegotthewordtogo.
Themodelhaspredictedcarcrimeinaboxontheirbeat.
Youknowwhatthecallisright?
Thefiveincidentsweremacerated.
Lookslikeit'
sakid.
Yeah,it'
sthesameaddressasthatkidthatwehadyesterday.
Suspectsfour,male...Whentheyreachtheirassignedhotspot,theyfindacold-platedcar.
Thelicenceplatesdon'
tmatchthevehicle.
regettingwhattheyneed,huh?
Whentheycallthenumberin,itturnsoutthecar'
sbeenstolen.
Itwasanareawherethere'
salotofGTAs,whichis"
Grandtheftauto"
peoplewerestealingcars.
Rightoutofrollcall,rightwhenwegotdownoneoftheboxestheywentinto,oneoftheareastheystartedpatrolling,rightawaytheyranacaranditcamebackstolen.
InFoothill,theyfoundusi