5.2.1 Galaxy Zoo

Galaxy Zoo anosanganisa kuitwa nevakawanda vakanga vasiri nyanzvi vanozvipira patsanura miriyoni enyeredzi.

Galaxy Zoo akakura kubva dambudziko nezvakaonekwa Kevin Schawinski, kudzidza mudzidzi ari Astronomy paUniversity of Oxford muna 2007. kurerutsa vakasvimha, Schawinski aifarira mapoka enyeredzi, uye mapoka enyeredzi zvinogona of nokuda Morphology-elliptical kana ravo zononoka-uye nokuda kwavo ruvara-mutema kana tsvuku. Panguva iyoyo, uchenjeri akawanda panyanzvi dzezvomuchadenga chaiva kuti zononoka mapoka enyeredzi, kufanana wedu Gwara Way, vakanga zvitema ruvara (zvinoratidza vechidiki) uye kuti elliptical mapoka enyeredzi vaiva yakatsvuka ruvara (zvinoratidza kuchembera). Schawinski vakakonona uchenjeri ichi akawanda. Iye vaifungira kuti apo muenzaniso uyu angava zvechokwadi zvavo, pane zvimwe zvakanga wandei kunze, uye kuti nokudzidza mijenya enyeredzi-ari kujairika ava havana kukodzera akatarisirwa muenzaniso-aigona kudzidza chinhu pamusoro muitiro kuburikidza iyo mapoka enyeredzi akaumba.

Saka, chii Schawinski raaida tichiputsa akawanda uchenjeri raiva guru akagadzirwa morphologically Classified enyeredzi; ndiko kuti, mapoka enyeredzi dzakanga semhosva kana zononoka kana elliptical. Dambudziko Zvisinei, aiva kuti huripo algorithmic Nzira kupatsanura vakanga vasati zvakanaka zvakakwana yaizoshandiswa kwesainzi; mune mamwe manzwi, classifying mapoka enyeredzi chaiva, panguva iyoyo, dambudziko raiva nesimba makombiyuta. Naizvozvo, chii chaidiwa hombe nhamba yevanhu of enyeredzi. Schawinski atanga dambudziko iri kupatsanura pamwe nechido kudzidza mudzidzi. In yedaro refu chegungano nomwe, 12-awa mazuva, akakwanisa patsanura 50,000 enyeredzi. Nepo 50,000 enyeredzi zvinganzwika sezvakawanda, zviri chaizvoizvo chete anenge 5% iri kusvika miriyoni imwe mapoka enyeredzi akanga mifananidzo iri Sloan Digital Sky Survey. Schawinski akaziva kuti aida zvikuru scalable nzira.

Sezvineiwo, zvinoitika kuti basa classifying enyeredzi haatarisiri kudzidzisa mazhinji nezvenyeredzi; unogona kudzidzisa munhu kuzviita nokukurumidza vakanaka. Nemamwe mashoko, kunyange zvazvo classifying mapoka enyeredzi ari basa rakanga rakaoma nokuti makombiyuta, zvakanga tsvarakadenga nyore kuti vanhu. Saka, apo akagara ari pub Oxford, Schawinski nevamwe zvomuchadenga Chris Lintott akarota imwe Website apo vazvipiri patsanura mifananidzo enyeredzi. Mwedzi mishomanana gare gare, Galaxy Zoo akaberekwa.

Panguva Galaxy Zoo Website, vazvipiri anozopiswa maminitsi mashoma kudzidziswa; Somuenzaniso, kudzidza musiyano zononoka uye elliptical renyeredzi (Figure 5.2). Shure nokudzidziswa uku, vazvipiri aifanira kupfuura munhu nyore eMibvunzo-nemazvo classifying 11 15 enyeredzi pamwe anozivikanwa classifications-uye zvadaro anozvipira yaizotanga kupatsanura yechokwadi vasingazivikanwi enyeredzi kuburikidza nyore web-inobva inowanikwa (Figure 5.3). Kuchinja kubva zvinozvipira wenyeredzi zvaizoitika mumaminitsi asingasviki 10 uye chete zvaida achipfuura akavaita zvipingamupinyi, nyore eMibvunzo.

Figure 5.2: Mienzaniso miviri Mhando huru enyeredzi: zononoka uye elliptical. The Galaxy Zoo Chirongwa akashandisa vazvipiri vanopfuura 100,000 kuti muzvikwata anopfuura 900,000 mifananidzo. Source: www.galaxyzoo.org .

Figure 5.3: Inzwa Zvinotaurwa chidzitiro apo vavhoti vakakumbirwa patsanura chifananidzo chete. Source: www.galaxyzoo.org .

Galaxy Zoo akakwezva rayo kutanga vazvipiri pashure basa racho anoratidzwa nhau nyaya, uye mwedzi inenge mitanhatu chirongwa rakakura zvinopfuura 100,000 mugari masayendisiti, vanhu vakanga aimbotaurawo nokuti Akanakidzwa basa uye vaida kubatsira kufanoziva zvenyeredzi. Pamwe chete, izvi Vazvipiri 100,000 vakabatsira dzinosvika vanopfuura mamiriyoni 40 classifications, pamwe vakawanda classifications vachibva zviduku, makirisito boka vechikamu (Lintott et al. 2008) .

Vatsvakurudzi vane ruzivo vakabhadhara Rokutenderwa tsvakurudzo vabatsiri kuti pakarepo vasinganyatsozvibvumi pamusoro date unhu. Nepo kusava nechokwadi ichi anonzwisisa, Galaxy Zoo inoratidza kuti kana rokuzvipira mipiro vari nemazvo kucheneswa, debiased, uye aggregated, vanogona kubereka yepamusoro zvabuda (Lintott et al. 2008) . An chinokosha Wandinyengedzerei nokuda kuwana chaunga kuti kusika nyanzvi unhu Data ndiye redundancy; kureva wakati chete basa kwakaitwa zhinji vanhu vakasiyana. Muna Galaxy Zoo, paiva vanenge 40 classifications por renyeredzi; Vanotsvakurudza vachishandisa Rokutenderwa tsvakurudzo vabatsiri zvaisazofa mari iyi pamwero redundancy uye naizvozvo tinofanira kuva zvikuru nezvokuita kunaka mumwe nomumwe kupatsanura. Chii vazvipiri waishaiwa mukurovedza, vakaita nokuda chete redundancy.

Kunyange multiple classifications por renyeredzi Zvisinei, mubatanidzwa yakatarwa kuzvipira classifications kubudisa tabvumirana kupatsanura panonetsa. Nekuti nematambudziko akafanana chaizvo kuchimuka dzakawanda dzevanhu computation zvirongwa zvinobatsira muchidimbu matanho matatu kuti Galaxy Zoo vatsvakurudzi vaishandisa kubereka ravo tenderano classifications. Chokutanga, vatsvakurudzi "vakachenesa" humboo nokubvisa Bogus classifications. Somuenzaniso, vanhu avo kakawanda of chete renyeredzi-chinhu zvaizoitika kudai vakanga vachiedza kukorovhera zvabuda-akanga classifications zvavo zvose kuraswa. Izvi nezvimwe kuchenesa chakafanana akabvisa anenge 4% yevanhu classifications zvose.

Chechipiri, mushure rokuchenesa, vaongorori vaifanira kubvisa enderana vakanga vaine kwavakarerekera mune classifications. Kuburikidza dzakatevedzana zvidzidzo rerekero vasabatwa midzi mukati chepakutanga chirongwa-somuenzaniso, tichiratidza vamwe vazvipiri pakati renyeredzi mune monochrome pachinzvimbo mavara-vaongorori vakaona akawanda enderana vakanga vaine kwavakarerekera, zvakadai somunhu hurongwa rerekero kuti patsanura kure zononoka mapoka enyeredzi sezvo elliptical mapoka enyeredzi (Bamford et al. 2009) . Kugadzirisa izvi hurongwa vakanga vaine kwavakarerekera Zvinokosha nokuti avhareji mipiro yakawanda hakuiti kubvisa enderana kusasarura; chete anobvisa kukanganisa chero.

Pakupedzisira, pashure debiasing, vaongorori vaida nzira chokubatanidza munhu classifications kubudisa tabvumirana kupatsanura. Nzira iri nyore chokubatanidza classifications nechimwe renyeredzi zvaizova kusarudza inonyanya kupatsanura. Zvisinei, nzira iyi aizopa mumwe muzvipiri wakaenzana uremu, uye vaongorori vaifungira kuti vamwe vazvipiri zvaiva nani pane kupatsanura pane vamwe. Saka, vatsvakurudzi yakatanga zvikuru kunzwisisa iterative weighting nzira kuti unoedza vapiwa kuona zvakanaka classifiers uye kuvapa mamwe uremu.

Saka, pashure zvitatu danho muitiro-kuchenesa, debiasing, uye weighting-the Galaxy Zoo tsvakurudzo chikwata vakanga vatendeukira rokuzvipira classifications mamiriyoni 40 vakapinda kutaridza nekubvirana morphological classifications. Kana izvi Galaxy Zoo classifications vakanga zvichienzaniswa nhatu yapfuura maduku-dzinotambudza pakuedza vanoongorora nyeredzi kwounyanzvi, kusanganisira kupatsanura kubudikidza Schawinski kuti zvakabatsira kuti afuridzire Galaxy Zoo, pakanga chibvumirano chakasimba. Saka, vazvipiri, muna uwandu hwezvinhu zvose zvabatanidzwa., vakakwanisa kupa yepamusoro classifications uye chikero kuti vaongorori kugona kunowirirana (Lintott et al. 2008) . Kutaura zvazviri, nokuva classifications dzevanhu chakadaro vazhinji enyeredzi, Schawinski, Lintott, uye vamwe vaigona kuratidza kuti chete vanenge 80% enyeredzi vanotevedzera kutarisirwa muenzaniso-yebhuruu spirals uye tsvuku ellipticals-uye vakawanda mapepa zvakanyorwa pamusoro kuwanikwa (Fortson et al. 2011) .

Akapiwa ichi nechekumashure, tinogona zvino kuona sei Galaxy Zoo anotevera vanorambana kushandisa-sangana yokubika, zvakafanana yokubika rinoshandiswa vakawanda kwevanhu computation zvirongwa. Chokutanga, dambudziko guru riri rwakakamukana akakora. Panyaya iyi, dambudziko classifying miriyoni mapoka enyeredzi ari rwakakamukana kuva miriyoni zvinetso classifying mumwe renyeredzi. Zvadaro, kuvhiyiwa rinoshandisirwa mumwe bimvu tisingatungamirirwi. Panyaya iyi, mumwe muzvipiri chaizoita patsanura mumwe renyeredzi sezvo kana zononoka kana elliptical. Pakupedzisira, migumisiro ikashandiswa kubudisa tabvumirana mugumisiro. Panyaya iyi, iyo nokubatanidza danho vaisanganisira rokutsvaira, debiasing, uye weighting kubereka tabvumirana kupatsanura neimwe renyeredzi. Kunyange zvazvo mabasa vakawanda vanoshandisa ichi zvayo nzira, mumwe nomumwe matanho anofanira customized kuti yakananga dambudziko riri kugadziriswa. Somuenzaniso, muna computation kwevanhu chirongwa inorondedzerwa pazasi, ndiye yokubika ichateverwa, asi kushandisa nokubatanidza matanho zvichava zvakasiyana.

Nokuti Galaxy Zoo chikwata, chirongwa chekutanga ichi kwaingova kutanga. Nokukurumidza vakaziva kuti kunyange zvazvo vakakwanisa patsanura pedyo miriyoni mapoka enyeredzi, pamwero uyu hazvina kukwana kushanda kuvatsva digitaalinen denga kuongorora, izvo zvinogona kubudisa mifananidzo inenge mabhiriyoni 10 mapoka enyeredzi (Kuminski et al. 2014) . Kubata kuwedzera kubva 1 miriyoni kusvika 10 mabhiriyoni-chinhu chinokosha pamusoro 10,000-Galaxy Zoo aifanira rokutsvaka nehasha 10,000 nguva yakawanda vechikamu. Kunyange zvazvo nhamba yevazvipiri paIndaneti guru, hazvisi haaperi. Saka, vatsvakurudzi vakaona kuti kana vari kuenda kubata nokusingaperi kukura uwandu data, itsva, kunyange kupfuura scalable, nzira yaidiwa.

Naizvozvo, Manda Banerji-vachishanda Kevin Schawinski, Chris Lintott, uye dzimwe nhengo Galaxy Zoo chikwata-kutanga dzidziso makombiyuta patsanura enyeredzi. More zvakananga, achishandisa vanhu classifications vakasikwa Galaxy Zoo, Banerji et al. (2010) akavaka muchina pakudzidza womuenzaniso aigona kufanotaura yevanhu kupatsanura munhu renyeredzi inobva zvechadenga mufananidzo. Kana ichi muchina kudzidza muenzaniso aigona obudisa dzevanhu classifications vaine zvakarurama, saka aigona kushandiswa Galaxy Zoo vatsvakurudzi kuti patsanura ane vanotova dzisingaperi nhamba enyeredzi.

Chinoumba Banerji uye nevamwe 'wokuita chaizvoizvo runako dzakafanana unyanzvi yaiwanzoshandiswa evanhu tsvakurudzo, kunyange zvazvo kuti kufanana irege kuva akajeka pakutanga kuaona. Chokutanga, Banerji uye nevamwe vagotendeuka mufananidzo nerimwe rive kutaridza Numeric zvinhu muchidimbu zviri ehupfumi. Somuenzaniso, mifananidzo enyeredzi pangava zvitatu zvakakanganisika; uwandu zvitema mumufananidzo, uye kutsaura mukati kupenya pixels, uye kuwanda pixels dzisiri chena. Kusarudzwa yakarurama zvinhu chinhu chinokosha dambudziko, uye kazhinji zvinoda pasi-nharaunda unyanzvi. Danho iri rokutanga, anowanzonzi kuchiitwa engineering, zvinoguma date chizvaro pamwe musara imwe mufananidzo uye ipapo matatu mbiru kurondedzera mufananidzo kuti. Apiwa date chizvaro uye vaishuva goho (semuenzaniso, kana mufananidzo yakatorwa semhosva achishandisa munhu sechinhu elliptical renyeredzi), nevatsvakurudzi inofungidzira ari parameters ane nhamba muenzaniso-somuenzaniso, chinhu chakafanana logistic regression-kuti inofanotaura ari kupatsanura munhu kwakavakirwa pamusoro zviri mufananidzo. Pakupedzisira, nevatsvakurudzi anoshandisa parameters iyi nhamba muenzaniso kuti pave vanofungidzirwa classifications enyeredzi matsva (Figure 5.4). Kufunga pasocial analoginen, fungidzira kuti iwe wakange wevanotungamirira mashoko pamusoro vadzidzi mamiriyoni, uye unoziva kana kudzidza pakoreji kana kwete. Unogona anokodzera ane logistic regression nemashoko ichi, uye ipapo unogona kushandisa kunoguma muenzaniso parameters kufanotaura kana vadzidzi vatsva vari kuenda kupedza kukoreji. In muchina kudzidza, ichi nzira-kushandisa akapomerawo mienzaniso kuumba uwandu muenzaniso kuti ungapa tarambana itsva mashoko-anonzi akatarisira kudzidza (Hastie, Tibshirani, and Friedman 2009) .

Figure 5.4: Simplified rondedzero sei Banerji et al. (2010) akashandisa Galaxy Zoo classifications kurovedza muchina pakudzidza muenzaniso kuita renyeredzi kupatsanura. Images enyeredzi vakanga vatendeuka ari chizvaro pakati zvinhu. In muenzaniso uyu Yokudzidza pane zvinhu zvitatu (uwandu zvitema mumufananidzo, uye kutsaura mukati kupenya pixels, uye kuwanda pixels dzisiri chena). Zvadaro, nokuti subset pakati zvifananidzo, iyo Galaxy Zoo pamapepa aya anoshandiswa kurovedza muenzaniso muchina pakudzidza. Pakupedzisira, muchina pakudzidza inoshandiswa kufungidzira classifications nokuda mapoka enyeredzi yasara. Ndinodana basa mhando iyi yechipiri-chizvarwa dzevanhu computational chirongwa nokuti, pane kuva vanhu kugadzirisa dambudziko, vane vanhu kuvaka dataset inogona kushandiswa kurovedza kombiyuta yokugadzirisa dambudziko. Kunakisa kombiyuta-vaibatsira ichi kusvika ndechokuti inoita kuti kubata vanotova parinogumira rakawandisa mashoko uchishandisa chete anokamuka uwandu nesimba kwevanhu.

The Mumagazini Banerji et al. (2010) muchina kudzidza womuenzaniso akanga akaoma kupfuura vaya wokutamba rangu muenzaniso-somuenzaniso, akashandisa zvinhu se "of Vaucouleurs kukodzera axial reshiyo" -uye muenzaniso wake akanga asina logistic regression, yaiva hazvizivi Neural network. Uchishandisa zvinhu kwake, muenzaniso, uye nekubvumirana Galaxy Zoo classifications, akakwanisa kusika Zviidzo imwe neimwe inoti, uye ipapo kushandisa Zviidzo izvi kuti zvakafanotaurwa pamusoro kupatsanura enyeredzi. Somuenzaniso, ongororo wake akawana kuti mifananidzo pasi "of Vaucouleurs kukwanawo axial reshiyo" vakanga anowanzova zononoka enyeredzi. Vakapiwa Zviidzo izvi, akakwanisa kufungidzira vanhu kupatsanura pakati renyeredzi vanonzwisisa zvakarurama.

Basa Banerji et al. (2010) akashandura Galaxy Zoo kupinda zvandakanga aizotumidza wechipiri-chizvarwa dzevanhu computation hurongwa. Nzira yakanakisisa kufunga izvi yechipiri-chizvarwa hurongwa ndechokuti pane kuva vanhu kugadzirisa dambudziko, vane vanhu kuvaka dataset inogona kushandiswa kurovedza kombiyuta yokugadzirisa dambudziko. Huwandu mashoko vaifanira kudzidzisa kombiyuta inogona kuva akakura zvokuti rinoda vakawanda womunhu kutsigirana kuti kusika. Panyaya Galaxy Zoo, ari Neural network rakashandiswa Banerji et al. (2010) waida yakakura chaizvo nhamba mienzaniso akapomerawo yevanhu-kuitira kuvaka muenzaniso kuti aikwanisa yakavimbika obudisa dzevanhu kupatsanura.

Kunakisa kombiyuta-vaibatsira ichi kusvika ndechokuti inoita kuti kubata vanotova parinogumira rakawandisa mashoko uchishandisa chete anokamuka uwandu nesimba kwevanhu. Somuenzaniso, mumwe mutsvakurudzi ane miriyoni evanhu of mapoka enyeredzi ungavaka chokufenbera muenzaniso kuti ipapo hungashandiswa patsanura bhiriyoni kana kunyange tiririyoni enyeredzi. Kana pane zvikuru nhamba emapoka enyeredzi, ipapo kwevanhu-kombiyuta chemasanganiswa urwu ndiyo chete Zvingangokubatsirai chaizvo. Uyu magumo scalability haasi kusununguka Zvisinei. Kuvaka muchina pakudzidza womuenzaniso kuti zvakarurama vanogona obudisa dzevanhu classifications pachayo dambudziko rakaoma, asi sezvinei varipo kare mabhuku yakanakisisa yakatsaurirwa nyaya iyi (Hastie, Tibshirani, and Friedman 2009; Murphy 2012; James et al. 2013) .

Galaxy Zoo rinoratidza Kushanduka computation dzakawanda dzevanhu zvirongwa. Kutanga, mumwe mutsvakurudzi kuedza chirongwa oga kana aine chikwata chidiki kutsvakurudza vabatsiri (semuenzaniso, Schawinski atanga kupatsanura nesimba). Kana nzira iri haasi kukwira zvakanaka, nevatsvakurudzi vanogona kuenda computation hwevanhu chirongwa apo vanhu vakawanda vanobatsira classifications. Asi, kwenguva yakati Vhoriyamu data, vakachena kuedza kwevanhu haangakwaniri. Panguva iyoyo, vatsvakurudzi vanofanira kuvaka wechipiri-chizvarwa gadziriro apo classifications evanhu inoshandiswa kudzidzisa muchina pakudzidza womuenzaniso kuti ipapo zvinogona kushandiswa husingakuvadzi risingaperi rakawandisa mashoko.