5.2.1 Galaxy Zoo
Galaxy Zoo combines lub zog ntawm ntau uas tsis yog-kws muaj txuj tuaj pab dawb cais ib tug lab galaxies.
Galaxy Zoo loj hlob tawm ntawm ib qho teeb meem kev nyuaj siab los ntawm Kevin Schawinski, ib tug kawm tiav me nyuam kawm ntawv nyob rau hauv Astronomy nyob rau University of Oxford nyob rau hauv 2007. simplifying ib tug heev me ntsis, Schawinski yog txaus siab nyob rau hauv galaxies, thiab galaxies yuav txwv kom muab zais los ntawm lawv cov morphology-elliptical los yog muab kauv-thiab los ntawm lawv cov xim-xiav los yog xim liab. Thaum lub sij hawm, pa txhab nrog astronomers yog tias muab kauv galaxies, zoo li peb kab lig ntuj txoj kev, twb xiav nyob rau hauv cov xim (qhia cov hluas) thiab hais tias elliptical galaxies twb liab nyob rau hauv cov xim (qhia cov laus). Schawinski ua xyem xyav no pa txawj ntse. Nws xav tias hais tias thaum no txawv tej zaum yuav muaj tseeb nyob rau hauv Feem ntau, muaj tej zaum ib tug loj tsawv naj npawb ntawm cov kev zam, thiab hais tias los ntawm kev kawm ntau ntawm cov txawv txawv galaxies-cov uas tsis haum cov kev tsim cov qauv-nws yuav kawm ib yam dab tsi txog tus txheej txheem los ntawm kev uas galaxies tsim.
Yog li, yog dab tsi Schawinski yuav tsum tau nyob rau hauv thiaj li yuav ntxeev pa txhab yog ib tug loj set ntawm morphologically classified galaxies; uas yog, galaxies uas tau dej num raws li yog muab kauv los yog elliptical. Qhov teeb meem, txawm li cas los, yog hais tias uas twb muaj lawm algorithmic txoj kev rau kev sib faib tseem tsis tau zoo txaus yuav tsum tau siv rau kev tshawb fawb; nyob rau hauv lwm yam lus, classifying galaxies yog, nyob rau ntawm lub sij hawm, ib tug teeb meem uas yog ib qho nyuaj rau computers. Yog li ntawd, dab tsi twb yuav tsum tau yog ib tug loj tus naj npawb ntawm tib neeg txwv kom muab zais galaxies. Schawinski undertook no faib teeb meem nrog tus yam txaus siab ntawm ib tug kawm tiav me nyuam kawm ntawv. Nyob rau hauv ib marathon sib kho ntawm xya, 12-teev hnub, nws muaj peev xwm cais 50,000 galaxies. Thaum 50,000 galaxies tej zaum yuav lub suab zoo li ib tug ntau, nws yog ua tau tsuas yog hais txog 5% ntawm cov yuav luag ib lab galaxies uas tau yees duab nyob rau hauv lub Sloan Digital Ntuj Survey. Schawinski pom tau hais tias nws xav tau ib tug ntau scalable mus kom ze.
Qhov zoo ces, nws hloov tawm hais tias cov hauj lwm ntawm classifying galaxies tsis yuav tsum tau tshaj kev kawm nyob rau hauv astronomy; koj muaj peev xwm qhia ib tug neeg ua nws zoo nkauj sai sai. Nyob rau hauv lwm yam lus, txawm tias classifying galaxies yog ib tug ua hauj lwm uas yog ib qho nyuaj rau computers, nws yog zoo nkauj yooj yim rau tib neeg. Yog li ntawd, thaum zaum nyob rau hauv ib tug pub nyob rau hauv Oxford, Schawinski thiab cov phooj astronomer Chris Lintott ua npau suav li ib lub website uas ua haujlwm pab dawb yuav faib cov dluab ntawm galaxies. Ib tug ob peb lub hlis tom qab ntawd, Galaxy Zoo yug.
Thaum lub Galaxy Zoo website, ua haujlwm pab dawb yuav undergo ib tug ob peb feeb ntawm kev kawm; piv txwv li, kev kawm sib txawv ntawm ib tug kauv thiab elliptical galaxy (Daim duab 5.2). Tom qab qhov kev kawm no, cov tuaj pab dawb yuav tsum tau dhau ib tug kuj yooj yim xeem ntawv-kom classifying 11 ntawm 15 galaxies nrog paub classifications-thiab ces tus ua haujlwm pab dawb yuav pib tiag tiag kev faib tsis paub hais tias galaxies los ntawm ib tug yooj yim web-based interface (Daim duab 5.3). Cov kev hloov ntawm kev pab dawb rau astronomer yuav siv sij hawm qhov chaw nyob rau hauv tsawg tshaj li 10 feeb thiab tsuas yog yuav tsum tau dua tus nqi qis tshaj ntawm hurdles, ib tug yooj yim kev xeem ntawv.
Daim duab 5.2: Piv txwv ntawm ob lub ntsiab hom ntawm galaxies: kauv thiab elliptical. Lub Galaxy Zoo project siv ntau tshaj 100,000 tuaj pab dawb rau pawg ntau tshaj 900.000 dluab. Tau qhov twg los: www.galaxyzoo.org .
Daim duab 5.3: tswv yim npo nyob qhov twg cov neeg xaiv tsa tau hais cais ib tug tib duab. Tau qhov twg los: www.galaxyzoo.org .
Galaxy Zoo attracted nws pib ua haujlwm pab dawb tom qab lub project twb featured nyob rau hauv ib tug xov xwm tsab xov xwm, thiab nyob rau hauv rau lub hlis hauv peb tes num loj hlob los mus koom nrog ntau tshaj 100,000 tus pej xeem zaum, cov neeg uas tau koom tes vim hais tias lawv nyiam cov neeg ua hauj lwm thiab lawv xav tau kev pab ua ntej astronomy. Ua ke, cov 100,000 tuaj pab dawb pab ib tug tag nrho ntawm ntau tshaj 40 lab classifications, nrog rau feem ntau ntawm cov classifications los ntawm ib tug kuj me me, cov tub ntxhais pab pawg neeg ntawm cov neeg (Lintott et al. 2008) .
Soj ntsuam uas muaj kev ntiav neeg undergraduate kev tshawb fawb pab tej zaum yuav tam sim ntawd yuav tsis ntseeg txog cov ntaub ntawv zoo. Thaum no skepticism yog tsim nyog, Galaxy Zoo qhia tau hais tias thaum ua haujlwm pab dawb pab yog raug ntxuav, debiased, thiab aggregated, lawv yuav ua tau high-quality tau (Lintott et al. 2008) . Ib qho tseem ceeb ua kom yuam kev tau txais cov neeg coob coob los mus tsim kev zoo cov ntaub ntawv yog redundancy; uas yog, muaj tib neeg ua hauj lwm ua los ntawm ntau cov neeg sib txawv. Nyob rau hauv Galaxy Zoo, muaj txog 40 classifications ib galaxy; soj ntsuam uas siv cov undergraduate kev tshawb fawb pab yuav tsis muaj nyiaj them rau theem no ntawm redundancy thiab yog li ntawd yuav tsum tau ntau npaum li cas kev txhawj xeeb nrog lub zoo ntawm txhua tus neeg cais. Yuav ua li cas tus tuaj pab dawb Corporate nyob rau hauv kev cob qhia, lawv ua los rau nrog redundancy.
Txawm muaj ntau yam classifications ib galaxy, txawm li cas los, sib txuas lub teeb ntawm pab dawb classifications yuav tsim ib tug kev pom kev faib yog lo qhia. Vim hais tias zoo heev txoj kev sib tw tshwm sim nyob rau hauv feem ntau tib neeg le caag tej yaam num, nws yog yuav pab tau rau luv luv saib xyuas peb cov kauj ruam uas lub Galaxy Zoo soj ntsuam siv los tsim lawv cov kev pom zoo classifications. Ua ntej, tus neeg tshawb fawb "ntxuav" cov ntaub ntawv los ntawm tshem bogus classifications. Piv txwv li, cov neeg uas pheej txwv kom muab zais rau hauv tib lub galaxy-ib yam dab tsi uas yuav tshwm sim yog hais tias lawv tau sim muab soj ntsuam-muaj tag nrho lawv tej classifications pov tseg. Qhov no thiab lwm yam zoo xws tu tshem tawm txog 4% ntawm tag nrho cov classifications.
Ob, tom qab tu, tus neeg tshawb fawb yuav tsum tau mus tshem tawm systematic biases nyob rau hauv classifications. Los ntawm ib tug series ntawm kev tsis ncaj ncees nrhiav kom tau cov kev tshawb fawb nyob ruaj nrees nyob rau hauv lub thawj project-piv txwv li, uas qhia ib co tuaj pab dawb cov galaxy nyob rau hauv monochrome es tsis txhob xim-kev soj ntsuam sab ob peb systematic biases, xws li ib tug systematic kev tsis ncaj ncees cais nyob deb deb kauv galaxies li elliptical galaxies (Bamford et al. 2009) . Kho rau cov systematic biases yog tsis tshua muaj ib qho tseem ceeb vim hais tias thaj ntau pab tsis tshem tawm systematic kev tsis ncaj ncees; nws tsuas tuskheej random kev ua yuam kev.
Thaum kawg, tom qab debiasing, tus neeg tshawb fawb yuav tsum tau ib txoj kev los muab tus neeg classifications yuav tsim ib tug kev pom kev faib. Qhov yooj yim txoj kev uas yuav muab classifications rau txhua galaxy yuav xaiv feem ntau cais. Txawm li cas los, qhov no mus kom ze yuav muab txhua pab dawb sib npaug zos luj, thiab cov soj ntsuam ntawm xav tias hais tias ib co tuaj pab dawb tau zoo dua kev faib dua lwm tus. Yog li ntawd, tus neeg tshawb fawb tsim ib tug ntau txoj iterative weighting txoj kev uas npaj siab mus txiav ntes qhov zoo tshaj plaws classifiers thiab muab rau lawv ntau dua ceeb thawj.
Yog li, tom qab ib tug peb kauj ruam cov txheej txheem-tu, debiasing, thiab weighting-lub Galaxy Zoo pab neeg tshawb fawb tau hloov dua siab tshiab 40 lab pab dawb classifications rau hauv ib lub txheej ntawm kev pom morphological classifications. Thaum cov Galaxy Zoo classifications tau muab piv rau peb yav dhau los me me-scale sim los ntawm kev astronomers, nrog rau cov kev faib los ntawm Schawinski uas pab txhawb Galaxy Zoo, muaj muaj zog kev pom zoo. Yog li, tus tuaj pab dawb, nyob hauv aggregate, twb tau muab kev zoo siab classifications thiab ib tug scale hais tias cov kev soj ntsuam yuav tsis phim (Lintott et al. 2008) . Nyob rau hauv qhov tseeb, yog muaj tib neeg classifications rau xws li ib tug loj tus naj npawb ntawm galaxies, Schawinski, Lintott, thiab lwm leej lwm tus twb tau mus qhia tias tsuas yog hais txog 80% ntawm cov galaxies ua raws li cov kev tsim cov qauv-xiav qwj thiab liab ellipticals-thiab heev heev cov ntaub ntawv tau raug sau txog no discovery (Fortson et al. 2011) .
Muab no, tom qab, peb muaj peev xwm tam sim no pom yuav ua li cas Galaxy Zoo raws li cov split-thov-muab daim ntawv qhia, cov tib daim ntawv qhia tias yog siv rau feem ntau tib neeg le caag txug tej yaam. Ua ntej, ib tug loj teeb meem yog phua rau hauv chunks. Nyob rau hauv cov ntaub ntawv no, qhov teeb meem ntawm classifying ib tug lab galaxies yog phua rau hauv ib tug lab cov teeb meem ntawm classifying ib galaxy. Tom ntej no, ib lub lag luam yog muaj ntaub ntawv rau txhua yam ntawm nws tus kheej. Nyob rau hauv cov ntaub ntawv no, ib tug volunteer yuav faib txhua galaxy li yog muab kauv los yog elliptical. Thaum kawg, qhov kev tshwm sim yog ua ke mus ua ib tug kev pom tshwm sim. Nyob rau hauv cov ntaub ntawv no, lub muab kauj ruam nrog cov tu, debiasing, thiab weighting yuav tsim ib tug kev pom kev faib rau txhua galaxy. Txawm tias feem ntau tej yaam num siv no xwb daim ntawv qhia, txhua ntawm cov kauj ruam yuav tsum tau Mekas rau cov kev teeb meem raug nyob. Piv txwv li, nyob rau hauv tus tib neeg lub le caag peb tes num hauv qab no, tib daim ntawv qhia yuav tsum tau ua raws, tab sis lub thov thiab muab cov kauj ruam yuav txawv heev.
Rau cov Galaxy Zoo pab neeg, qhov no ua ntej peb tes num yog nyuam qhuav pib. Heev sai sai lawv pom tau hais tias txawm lawv twb tau cais ze rau ib tug lab galaxies, qhov no scale yog tsis txaus mus ua hauj lwm nrog newer cov ntuj cov ntawv tshawb fawb, uas yuav tsim dluab ntawm txog 10 billion galaxies (Kuminski et al. 2014) . Mus leg ib qho kev nce ntawm 1 lab rau 10 billion-ib qhov zoo tshaj ntawm 10,000-Galaxy Zoo yuav tsum kom txhij roughly 10,000 lub sij hawm ntau tuaj koom. Txawm tias cov xov tooj ntawm tuaj pab dawb nyob rau hauv Internet yog loj, nws yog tsis mus ib txhis. Yog li ntawd, tus neeg tshawb fawb pom tau hais tias yog hais tias lawv yuav mus ko puas tau loj hlob nqi ntawm cov ntaub ntawv, ib tug tshiab, txawm ntau scalable, ze twb yuav tsum tau.
Yog li ntawd, Manda Banerji-ua hauj lwm nrog Kevin Schawinski, Chris Lintott, thiab lwm tus neeg lub Galaxy Zoo pab neeg no pib qhia computers cais galaxies. Dua, siv tus tib neeg classifications tsim los ntawm Galaxy Zoo, Banerji et al. (2010) ua ib tug tshuab kev kawm cov qauv uas yuav twv seb tus tib neeg kev faib ntawm ib tug galaxy raws li nyob rau hauv cov yam ntxwv ntawm cov duab. Yog hais tias qhov no tshuab kev kawm qauv yuav muaj me tub tus tib neeg classifications uas muaj tseeb, ces nws yuav raug siv los ntawm Galaxy Zoo soj ntsuam cais ib tug yeej tseem zoo infinite tus naj npawb ntawm galaxies.
Cov tub ntxhais ntawm Banerji thiab lug txhawb cov miv 'mus kom ze yog ua tau zoo nkauj zoo xws li cov hom kev kawm feem ntau siv nyob rau hauv kev sib raug zoo kev tshawb fawb, txawm hais tias hais tias kev zoo sib thooj yuav tsis muaj tseeb thaum xub thawj siab ib muag. Ua ntej, Banerji thiab lug txhawb cov miv hloov dua siab tshiab ib cov duab rau hauv ib lub txheej ntawm numeric nta uas saib lub ntsab lug nws yog lub zog. Piv txwv li, rau cov dluab ntawm galaxies muaj yuav ua tau peb nta: tus nqi ntawm cov kob xiav nyob rau hauv cov duab, tus variance nyob rau hauv lub qhov ci ntawm lub pixels, thiab cov kev faib ua feem ntawm uas tsis yog-dawb pixels. Cov kev xaiv ntawm qhov tseeb nta yog ib qho tseem ceeb ib feem ntawm qhov teeb meem, thiab nws feem ntau yuav tsum tau kawm-cheeb tsam kev txawj ntse. Qhov no yog thawj kauj ruam, feem ntau hu ua feature engineering, tau nyob rau hauv ib cov ntaub ntawv matrix nrog ib tug kab ib daim duab thiab ces peb txhua piav txog hais tias cov duab. Muab cov ntaub ntawv matrix thiab cov yam tso zis (xws li, seb tus duab twb txwv kom muab zais los ntawm ib tug tib neeg raws li ib tug elliptical galaxy), tus kws tshawb fawb kev kwv yees qhov tsis muaj ib tug statistical qauv-piv txwv li, ib yam dab tsi zoo li ib tug logistic regression-uas kwv yees rau cov tib neeg kev faib raws li ntawm cov yam ntxwv ntawm cov duab. Thaum kawg, tus kws tshawb fawb yuav siv cov tsis nyob rau hauv no statistical qauv tsim kwv yees li ntawm classifications tshiab galaxies (Daim duab 5.4). Mus xav seb ib tug social analog, xav txog tej yam uas koj muaj demographic ntaub ntawv hais txog ib tug lab cov tub ntxhais kawm, thiab koj paub hais tias lawv kawm tiav los ntawm tsev kawm qib siab los yog tsis. Koj yuav haum ib tug logistic regression rau cov ntaub ntawv no, thiab ces koj yuav tau siv lub resulting qauv ciaj ciam mus rau kwv yees seb menyuam kawm ntawv tshiab yog yuav mus kawm tiav los ntawm tsev kawm qib siab. Nyob rau hauv lub tshuab kev kawm, qhov no mus kom ze-siv sau tias piv txwv los ua ib tug statistical qauv uas yuav ces daim ntawv lo ntaub ntawv tshiab-yog hu ua saib xyuas kev kawm (Hastie, Tibshirani, and Friedman 2009) .
Daim duab 5.4: Yooj yim zog hauj lwm ntawm yuav ua li cas Banerji et al. (2010) siv lub Galaxy Zoo classifications los qhia ib lub tshuab kev kawm cov qauv los ua galaxy cais. Dluab ntawm galaxies tau hloov dua siab tshiab rau hauv ib lub matrix ntawm cov nta. Nyob rau hauv no yooj yim to taub Piv txwv tias muaj peb nta (tus nqi ntawm cov kob xiav nyob rau hauv cov duab, tus variance nyob rau hauv lub qhov ci ntawm lub pixels, thiab cov kev faib ua feem ntawm uas tsis yog-dawb pixels). Ces, rau ib tug subset ntawm cov dluab, lub Galaxy Zoo ntawv yog siv los cob qhia ib tug tshuab kev kawm qauv. Thaum kawg, lub tshuab kev kawm yog siv los laij cov nqi classifications rau hauv seem uas yog galaxies. Kuv hu qhov no zoo ntawm peb tes num ib tug thib ob-tiam tib neeg computational project vim hais tias, es muaj tib neeg daws tau ib qho teeb meem, lawv muaj tib neeg tsim ib tug dataset uas yuav siv tau los cob qhia ib lub computer los mus daws cov teeb meem. Qhov kom zoo dua ntawm no lub computer-pab mus kom ze yog hais tias nws enables koj yuav lis yeej tseem zoo infinite nqi ntawm cov ntaub ntawv siv xwb ib tug finite nyiaj ntawm tib neeg lub dag zog.
Cov yam ntxwv nyob rau hauv Banerji et al. (2010) tshuab kev kawm qauv ntau dua cov neeg nyob rau hauv kuv qho khoom ua si piv txwv-piv txwv li, nws siv nta xws li "de Vaucouleurs haum axial piv" -thiab nws qauv tsis logistic regression, nws yog ib tug dag neural network. Siv nws nta, nws qauv, thiab cov kev pom zoo Galaxy Zoo classifications, nws muaj peev xwm mus tsim tau tes taw hnyav li rau txhua feature, thiab ces siv tau cov no tes taw hnyav li yuav ua rau txoj hau kev yuav txog lub kev faib ntawm galaxies. Piv txwv li, nws tsom xam pom tias cov dluab nrog tsawg "de Vaucouleurs haum axial piv" yog feem ntau yuav tsum tau muab kauv galaxies. Muab cov tes taw hnyav li, nws muaj peev xwm mus twv seb tus tib neeg kev faib ntawm ib tug galaxy nrog tsim nyog raug.
Cov ua hauj lwm ntawm Banerji et al. (2010) muab Galaxy Zoo mus rau hauv dab tsi kuv yuav hu rau ib tug thib ob-tiam tib neeg le caag system. Qhov zoo tshaj plaws txoj kev uas yuav xav txog cov thib ob-tiam tshuab yog hais tias es tshaj muaj tib neeg daws tau ib qho teeb meem, lawv muaj tib neeg tsim ib tug dataset uas yuav siv tau los cob qhia ib lub computer los mus daws cov teeb meem. Tus nqi ntawm cov ntaub ntawv yuav tsum tau mus qhia rau hauv lub computer yuav tsum yog li ntawd loj hais tias nws yuav tsum tau ib tug tib neeg loj sib koom tes los ua. Nyob rau hauv cov ntaub ntawv ntawm Galaxy Zoo, cov neural tes hauj lwm siv los ntawm Banerji et al. (2010) yuav tsum tau ib tug heev loj tus naj npawb ntawm tib neeg uas sau tias piv txwv nyob rau hauv thiaj li yuav tsim kom muaj ib tug qauv uas twb tau mus nti me nyuam tib neeg cais.
Qhov kom zoo dua ntawm no lub computer-pab mus kom ze yog hais tias nws enables koj yuav lis yeej tseem zoo infinite nqi ntawm cov ntaub ntawv siv xwb ib tug finite nyiaj ntawm tib neeg lub dag zog. Piv txwv li, ib tug soj ntsuam nrog ib tug lab tib neeg txwv kom muab zais galaxies yuav tsim ib tug predictive qauv uas yuav ces yuav siv tau los cais ib tug billion los sis txawm ib trillion galaxies. Yog hais tias muaj yog enormous xov tooj ntawm galaxies, ces qhov no zoo ntawm tib neeg lub computer hybrid yog tiag tiag tsuas tau tshuaj. Qhov no infinite scalability yog tsis dawb, txawm li cas los. Lub tsev ib lub tshuab kev kawm cov qauv uas yuav raug me nyuam rau cov tib neeg classifications yog nws tus kheej ib tug muaj teeb meem, tab sis zoo hmoo muaj cov twb zoo heev phau ntawv nplooj siab rau qhov no lub ntsiab lus (Hastie, Tibshirani, and Friedman 2009; Murphy 2012; James et al. 2013) .
Galaxy Zoo qhia tau hais tias cov evolution ntawm ntau tib neeg le caag txug tej yaam. Ua ntej, ib tug soj ntsuam attempts lub project los ntawm nws tus kheej los yog nrog ib tug me me pab neeg ntawm cov kev tshawb fawb pab (xws li, Schawinski tus thawj zaug kev faib dag zog). Yog hais tias qhov no mus kom ze tsis scale zoo, tus kws tshawb fawb yuav tsiv mus nyob rau ib tug tib neeg le caag qhov project uas muaj coob tus neeg pab classifications. Tab sis, rau ib tug tej yam ntim ntawm cov ntaub ntawv, ntshiab tib neeg siab yuav tsis txaus. Nyob rau ntawm lub taw tes, soj ntsuam yuav tsum tau tsim ob-tiam systems uas tib neeg classifications yog siv los cob qhia ib tug tshuab kev kawm cov qauv uas yuav ces tau mus thov kom zoo unlimited nqi ntawm cov ntaub ntawv.
