5.2.1 Way Zoo

Galao Zoo inagwirizanitsa zoyesayesa za ambiri odzipereka osadziŵa kuti adziwe magulu a milalang'amba milioni.

Galao Zoo inachokera ku vuto lomwe Kevin Schawinski, wophunzira wophunzira wophunzira wophunzira wophunzira wophunzirayo anaphunzira ku yunivesite ya Oxford m'chaka cha 2007. Pang'ono ndi pang'ono, Schawinski anali ndi chidwi ndi milalang'amba, ndipo milalang'amba imatha kusindikizidwa ndi morphological-elliptical kapena spiral-and ndi mtundu wawo-buluu kapena wofiira. Panthawiyo, nzeru zodziwika bwino pakati pa akatswiri a sayansi ya zakuthambo zinali kuti milalang'amba yozungulira, monga Milky Way, inali ya mtundu wa buluu (yosonyeza achinyamata) ndi magalasi a elliptical anali ofiira (akusonyeza ukalamba). Schawinski anakayikira izi zachilendo nzeru. Anakayikira kuti ngakhale kuti chitsanzo ichi chikhoza kukhala chenicheni, mwina pangakhale nambala yochuluka kwambiri, ndipo kuti pophunzira milalang'amba yodabwitsayi-yomwe siinakwaniritsidwe ndi chizolowezi choyembekezeredwa-akhoza kuphunzira chinachake ponena za njira yomwe milalang'amba inakhazikitsidwa.

Motero, chimene Schawinski anachifuna pofuna kugonjetsa nzeru zachilendo chinali gulu lalikulu la milalang'amba yamagulu; ndiko kuti, milalang'amba yomwe idatchulidwa ngati yowonjezera kapena yotchedwa elliptical. Vuto, komabe, kuti njira zomwe zakhalapo kale zogwiritsira ntchito sizinagwiritsidwe ntchito pofufuza za sayansi; Mwa kuyankhula kwina, kugawa milalang'amba kunali, panthawiyo, vuto lomwe linali lovuta kwa makompyuta. Choncho chimene chinafunika ambiri nyenyezi anthu -classified. Schawinski adayambitsa vuto ili ndi chidwi cha wophunzira wophunzira. Mu gawo la marathon la masiku asanu ndi awiri (12) la ola limodzi, adatha kusankha magulu a magulu 50,000. Ngakhale milalang'amba 50,000 ingawoneke ngati yochuluka, kwenikweni ndi pafupi 5% ya milalang'amba pafupifupi milioni imodzi yomwe inafotokozedwa mu Sloan Digital Sky Survey. Schawinski anazindikira kuti anafunikira njira yowonjezereka.

Mwamwayi, likukhalira kuti ntchito mtundu nyenyezi sikutanthauza kukapitiriza maphunziro ku zakuthambo; mungaphunzitse munthu kuchita izo wokongola mwamsanga. M'mawu ena, ngakhale mtundu nyenyezi ndi ntchito lovuta kwa makompyuta, anali wokongola n'kovuta kwa anthu. Choncho, atakhala mu malo omwera mu Oxford, Schawinski ndi anzake zakuthambo Chris Lintott ndinalota mmwamba webusaiti kumene ongodzipereka m'kagulu zithunzi za nyenyezi. Patapita miyezi ingapo, Way Zoo anabadwa.

Pa webusaiti ya Galaxy Zoo, odzipereka amakhoza maphunziro ochepa; Mwachitsanzo, phunzirani kusiyana pakati pa galaxy yozungulira ndi yapamwamba (chithunzi 5.2). Pambuyo pa maphunzirowa, aliyense wodzipereka anayenera kudutsa mafunso ophweka a mitundu khumi ndi anayi ndi khumi ndi asanu ndi awiri (15) omwe ali ndi magulu odziwika bwino. Kusintha kuchokera kwa wodzipereka kupita kwa nyenyezi kudzachitika pasanathe mphindi 10 ndipo kungofunika kudutsa zovuta kwambiri, funso losavuta.

Chithunzi 5.2: Zitsanzo za mitundu ikuluikulu ya milalang'amba: kutuluka ndi elliptical. Ntchito ya Galaxy Zoo inagwiritsa ntchito odzipereka oposa 100,000 kuti agwirizanepo mafano oposa 900,000. Inaperekedwanso ndi chilolezo kuchokera ku http://www.GalaxyZoo.org ndi Sloan Digital Sky Survey .

Chithunzi 5.3: Pulogalamu yolembera yomwe anthu odzifunsira anafunsidwa kuti apange fano limodzi. Inaperekedwanso ndi chilolezo kuchokera ku Chris Lintott chozikidwa pa chithunzi kuchokera ku Sloan Digital Sky Survey .

Galao Zoo inakopeka odzipereka awo pokhapokha polojekitiyi inafotokozedwa m'nkhani yatsopano, ndipo pafupifupi miyezi isanu ndi umodzi polojekitiyi inakula ikuphatikizapo asayansi oposa 100,000, anthu omwe adagwira nawo ntchito chifukwa ankasangalala ndi ntchitoyi ndipo ankafuna kuthandiza chitukuko cham'tsogolo. Palimodzi, anthu odzipereka okwana 100,000 apereka ziwerengero zoposa 40 miliyoni, ndipo ambiri mwa magawowa amachokera ku gulu laling'ono, omwe ndi a (Lintott et al. 2008) .

Ochita kafukufuku amene akugwira ntchito yolemba kafukufuku wophunzira payekha angakhale osakayikira za khalidwe la deta. Ngakhale kukayikira kumeneku kuli koyenera, Galaxy Zoo amasonyeza kuti pamene zopereka zodzipatulira zitsukidwa bwino, zosasokonezeka, komanso zowonjezera, zingathe kupanga zotsatira zabwino kwambiri (Lintott et al. 2008) . Chinthu chofunika kwambiri chothandiza anthu kuti apange deta yamtengo wapatali ndi redundancy , ndiko kuti, kukhala ndi ntchito yomweyi yochitidwa ndi anthu osiyanasiyana. Mu Galaxy Zoo, panali madera pafupifupi 40 pa galaxy; ochita kafukufuku omwe amathandizira ochita kafukufuku wophunzira payekha sangakwanitse kupeza chiwerengero cha redundancy kotero kuti ayenera kukhala okhudzidwa ndi khalidwe la aliyense payekha. Chimene odzipereka sanaphunzitse, iwo anachipangira ndi redundancy.

Ngakhale kuti pali magulu angapo pamlalang'amba, komabe, kuphatikizapo magulu odzipereka kuti apange chigwirizano chinali chovuta. Chifukwa mavuto ambiri ofanana ndi omwe amapezeka m'mapulojekiti ambiri a anthu, ndizothandiza kufufuza mwachidule njira zitatu zomwe akatswiri a zofufuza za Zoo amagwiritsira ntchito kupanga chigwirizano chawo. Choyamba, ochita kafukufuku "adayeretsa" chidziwitsocho pochotsa zizindikiro zolakwika. Mwachitsanzo, anthu omwe amagawenga mlalang'amba yemodzi mobwerezabwereza-chinachake chimene chingachitike ngati akuyesera kuwonetsa zotsatira-anali atasintha zonse zomwe adaziyika. Ichi ndi zina zotere kuyeretsa zinachotsedwa pafupi ndi 4% mwazinthu zonse.

Chachiwiri, atatha kuyeretsa, ochita kafukufuku anafunika kuchotseratu zochitikazo. Kupyolera mu zochitika zotsatila zofufuza zomwe zili mkati mwa polojekiti yapachiyambi-mwachitsanzo, kusonyeza ena odzipereka mlalang'amba mu monochrome mmalo mwa mtundu-omwe ofufuza anapeza zingapo zowonongeka mwatsatanetsatane, monga chisokonezo chokhazikika kuti azigawa magalasi oyenda kutali monga magalasi a elliptical (Bamford et al. 2009) . Kukonzekera kuzinthu zowonongekazi ndikofunikira kwambiri chifukwa redundancy sichichotsa chisokonezo chenichenicho; Zimangothandiza kuthana ndi zolakwika zosavuta.

Pambuyo pake, atatha kukhumudwa, ochita kafukufuku anafuna njira yodziphatikiza ndi magulu osiyanasiyana kuti apange chigwirizano. Njira yosavuta yogwirizanitsa magawo a mlalang'amba uliwonse iyenera kuti ikhale yosankha mtundu wofala kwambiri. Komabe, njirayi ikanapangitsa aliyense wodzipereka wolemera wofanana, ndipo ochita kafukufuku akudandaula kuti ena odzipereka anali oposa bwino kuposa ena. Chifukwa chake, ochita kafukufukuyu anayamba njira yowonongeka yowonjezeramo kuti ayese kuyang'ana bwino kwambiri ndikuwapatsa kulemera kwake.

Choncho, patatha njira zitatu-kukonza, kuyerekezera, ndi kulemera-gulu la kafukufuku wa Galao lopangitsa kuti magulu 40 odzipereka aperekedwe kukhala chigwirizano cha morphological classifications. Pamene magulu a Galaxy Zoo anali oyerekeza ndi mayesero atatu apitalo ochepa a akatswiri a zakuthambo, kuphatikizapo kugawa kwa Schawinski komwe kunathandiza kulimbikitsa Galaxy Zoo, panali mgwirizano wamphamvu. Choncho, odzipereka, palimodzi, adatha kupereka zigawo zapamwamba komanso pazomwe ophunzirawo sangafanane nazo (Lintott et al. 2008) . Ndipotu, pokhala ndi magulu ambiri a milalang'amba, Schawinski, Lintott, ndi ena adatha kusonyeza kuti ndi magulu oposa 80% okha omwe amatsatira chitsanzo cha mtundu wa buluu ndi zofiira zofiira-ndipo mapepala ambiri alembedwa kupeza izi (Fortson et al. 2011) .

Kuchokera kumbuyoku, mukutha kuona momwe Zaka Zoo zimatsatilira chophatikizira-chophatikiza-chophatikizira, njira yomweyi yomwe imagwiritsidwa ntchito pamapangidwe ambiri a anthu. Choyamba, vuto lalikulu limagawidwa kukhala chunks. Pachifukwa ichi, vuto la kusankha magulu a milalang'amba milioni linagawanika kuti likhale milioni imodzi yogawa mlalang'amba umodzi. Kenaka, opaleshoni imagwiritsidwa ntchito pa chunk iliyonse mosiyana. Pachifukwa ichi, odzipereka amagawira mlalang'amba uliwonse ngati kuwombera kapena kuwombera. Pomalizira, zotsatira zimagwirizanitsidwa kuti zibweretse chigwirizano. Pachifukwa ichi, gawo lophatikizapo ndikuphatikizapo kuyeretsa, kupusitsa, ndi kulemetsa kuti apange chigwirizano cha magulu. Ngakhale kuti mapulogalamu ambiri amagwiritsira ntchito njirayi, gawo lililonse liyenera kusinthidwa kuti likhale lovuta. Mwachitsanzo, mu polojekiti yaumunthu yomwe ikufotokozedwa pansipa, njira yomweyi idzatsatiridwa, koma kugwiritsa ntchito ndikuphatikizapo njira zidzakhala zosiyana.

Kwa gulu la Galao Zoo, polojekiti yoyambayi inali chiyambi chabe. Mwamsanga anazindikira kuti ngakhale kuti adatha kugawira milalang'amba pafupi ndi milioni, izi sizingakwanitse kugwira ntchito ndi zofufuza zamakono zatsopano, zomwe zingapangitse zithunzi za magalasi pafupifupi 10 biliyoni (Kuminski et al. 2014) . Pofuna kuthandizira kuwonjezeka kwa 1 miliyoni kufika 10 biliyoni-chinthu cha Galaxy 10,000 chiyenera kuitanitsa anthu oposa 10,000. Ngakhale kuti chiwerengero cha odzipereka pa intaneti ndi chachikulu, sizitha. Chifukwa chake, ochita kafukufuku adazindikira kuti ngati adzalandira deta zambirimbiri, njira yatsopano, yosavuta kwambiri, ikanafunika.

Choncho, Manda Banerji-akugwira ntchito ndi Schawinski, Lintott, ndi ena a gulu la Galaxy Zoo (2010) -waphunzitsa makompyuta kuti azigawa magulu. Makamaka, pogwiritsa ntchito zida za anthu zomwe zinapangidwa ndi Galaxy Zoo, Banerji inapanga makina ophunzirira omwe angagwiritse ntchito mndandanda wa mtundu wa mlalang'amba wokhudzana ndi makhalidwe a fano. Ngati chithunzichi chikhoza kubweretsa zigawo zaumunthu molondola, ndiye kuti zikhoza kugwiritsidwa ntchito ndi ochita kafukufuku wa zolemba za Galaxy kuti awonetse milalang'amba yambiri yopanda malire.

Mfundo yaikulu ya Banerji ndi njira zomwe amagwira nawo ntchito ndizofanana kwambiri ndi njira zomwe zimagwiritsidwa ntchito pofufuza kafukufuku, ngakhale kuti kufanana sikungakhale koyambirira poyamba. Choyamba, Banerji ndi anzake atatembenuzidwa lililonse fano mu mndandanda wa zinthu n'zosangalatsa kumva kuti mwachidule katundu wake. Mwachitsanzo, pazithunzi za milalang'amba, pangakhale zinthu zitatu: kuchuluka kwa buluu mu chithunzi, kusiyana kwa kuwala kwa pixel, ndi chiwerengero cha pixels omwe sali oyera. Kusankhidwa kwa zinthu zofunikira ndi gawo lofunikira la vutoli, ndipo kawirikawiri limafuna luso la malo. Gawo loyambalo, lomwe nthawi zambiri limatchedwa kuti engineering engineering , limabweretsa chiwerengero cha deta ndi mzere umodzi pa fano ndiyeno zigawo zitatu zofotokoza fanolo. Chifukwa cha chiwerengero cha deta komanso zofunikirako zofunikira (mwachitsanzo, ngati chithunzicho chinayikidwa ndi munthu monga galaxy elliptical), wofufuzirayo amapanga chiwerengero cha masewero kapena makina ophunzitsira-mwachitsanzo, regression-logistic-yomwe inaneneratu chigawo chaumunthu chozikidwa pambaliyi za fano. Pomalizira, wofufuzirayo amagwiritsa ntchito njirayi kuti awonetse kuti pali milalang'amba yatsopano (Fanizo 5.4). Mu kuphunzira kwa makina, njirayi-kugwiritsa ntchito zitsanzo zolembedwa kuti apange chitsanzo chomwe chingathe kulemba deta yatsopano-imatchedwa kuphunzira oyang'aniridwa .

Chithunzi 5.4: Kufotokozedwa kwa Banerji et al. (2010) adagwiritsa ntchito zolemba za Galaxy Zoo kuti aphunzitse njira yophunzirira makina kuti apange magulu a magalasi. Zithunzi za milalang'amba zinatembenuzidwa mu matrix of features. Mu chitsanzo chophweka ichi, pali zigawo zitatu (kuchuluka kwa buluu mu chithunzi, kusiyana kwa kuwala kwa pixel, ndi chiwerengero cha pixels oswhite). Kenaka, polemba zithunzi, malemba a Galao Zoo amagwiritsidwa ntchito pophunzitsa makina ophunzirira. Pomalizira, makina opanga makinawa amagwiritsidwa ntchito kulingalira zolemba za magulu otsalawo. Ndikutcha kuti pulojekiti yothandizira anthu pakompyuta chifukwa, m'malo mokhala ndi anthu kuthetsa vuto, anthu adzipanga deta yomwe ingagwiritsidwe ntchito kuphunzitsa kompyuta kuthetsa vutoli. Ubwino wa pulogalamuyi yothandizidwa ndi makompyuta ndikuti zimakuthandizani kuti muzitha kugwiritsira ntchito deta yopanda malire pogwiritsa ntchito khama lambiri la anthu. Zithunzi za milalang'amba zimaperekedwa ndi chilolezo kuchokera ku Sloan Digital Sky Survey .

Zomwe zinali mu Banerji ndi machitidwe ogwiritsira ntchito makina zinali zovuta kwambiri kuposa zitsanzo zanga zogwiritsira ntchito - mwachitsanzo, amagwiritsa ntchito zinthu monga "de Vaucouleurs zoyenera". Pogwiritsira ntchito zida zake, chitsanzo chake, ndi kugwirizana kwa Galaxy Zoo classifications, iye adatha kupanga zolemera pa mbali iliyonse, ndiyeno amagwiritsa ntchito zolemera kuti aneneratu za mndandanda wa milalang'amba. Mwachitsanzo, kufufuza kwake kunapeza kuti mafano omwe ali otsika "a Vaucouleurs amagwirizana ndi axial chiŵerengero" anali otheka kukhala magulu a mizimu. Chifukwa cha zolemera zimenezi, adatha kufotokozera mtundu wa anthu wa galaxy molondola.

Ntchito ya Banerji ndi ogwira ntchito inasintha Galao Zoo m'zinthu zomwe ndikanati ndiyese njira yowunikira anthu . Njira yabwino yoganizira za machitidwe awa ndikuti m'malo mokhala ndi anthu kuthetsa vuto, iwo ali ndi anthu omwe amapanga dataset yomwe ingagwiritsidwe ntchito kuphunzitsa kompyuta kuthetsa vutoli. Nthawi zina, kuphunzitsa kompyuta kuthana ndi vutoli kungakhale ndi zitsanzo zambiri, ndipo njira yokhayo yoperekera zitsanzo zowonjezera ndi kugwirizana kwakukulu. Ubwino wa njirayi yothandizidwa ndi makompyuta ndikuti zimakuthandizani kuti muzitha kugwiritsira ntchito deta yopanda malire pogwiritsa ntchito kuchuluka kwa mphamvu zaumunthu. Mwachitsanzo, kafukufuku amene ali ndi milalang'amba ya anthu mamiliyoni ambiri akhoza kupanga njira yowonongeka yomwe ingagwiritsidwe ntchito kupanga magulu oposa biliyoni kapena trillion. Ngati pali milalang'amba yochulukirapo, ndiye kuti mtundu uwu wa anthu-wosakanizidwa ndi makompyuta ndiwo wokha womwe ungathetsere yankho. Kulephera kwake kosatha sikunali kwaulere, komabe. Kumanga makina ophunzirira makina omwe angabweretse bwino zigawo zaumunthu ndizovuta, koma mwachisangalalo pali mabuku abwino kwambiri operekedwa ku mutuwu (Hastie, Tibshirani, and Friedman 2009; Murphy 2012; James et al. 2013) .

Galao Zoo ndi chitsanzo chabwino cha kusintha kwa mapulani a anthu. Choyamba, wofufuza amayesa ntchitoyo yekha kapena gulu laling'ono la othandizira kafukufuku (mwachitsanzo, ntchito yoyamba ya Schawinski). Ngati njirayi isapitirire bwino, wofufuzirayo akhoza kusamukira ku polojekiti ya anthu ndi ophunzira ambiri. Koma, chifukwa cha chiwerengero china cha deta, khama laumunthu loyera silidzakwanira. Panthawi imeneyo, ochita kafukufuku amafunika kukhazikitsa njira yothandizira anthu omwe amagwiritsidwa ntchito ndi makompyuta komwe anthu amagwiritsidwa ntchito pophunzitsa makina ophunzirira omwe angagwiritsidwe ntchito pamtundu uliwonse wa deta.