Okuqukethwe[Fihla][Bonisa]
Imidlalo yevidiyo iyaqhubeka nokunikeza inselele izigidigidi zabadlali emhlabeni jikelele. Kungenzeka awukakwazi okwamanje, kodwa ama-algorithms wokufunda komshini aseqalile ukukhuphukela enseleleni nawo.
Njengamanje kunenani elibalulekile locwaningo emkhakheni we-AI ukubona ukuthi izindlela zokufunda zomshini zingasetshenziswa yini emidlalweni yevidiyo. Intuthuko enkulu kulo mkhakha ikhombisa lokho ukufunda imishini ama-agent angasetshenziselwa ukulingisa noma ngisho nokushintsha umdlali ongumuntu.
Kusho ukuthini lokhu ngekusasa lika amageyimu evidiyo?
Ingabe la maphrojekthi awokuzijabulisa nje, noma kukhona izizathu ezijulile ezenza abacwaningi abaningi bagxile emidlalweni?
Lesi sihloko sizohlola kafushane umlando we-AI emidlalweni yevidiyo. Ngemva kwalokho, sizokunikeza ukubuka konke okusheshayo kwamanye amasu okufunda omshini esingawasebenzisa ukuze sifunde ukushaya imidlalo. Sizobe sesibheka ezinye izinhlelo zokusebenza eziyimpumelelo ze amanetha e-neural ukuze ufunde futhi ubambe kahle imidlalo ethile yevidiyo.
Umlando omfushane we-AI kumageyimu
Ngaphambi kokuthi singene kokuthi kungani amanetha e-neural abe i-algorithm efanelekile yokuxazulula imidlalo yevidiyo, ake sibheke kafushane ukuthi ososayensi bamakhompiyutha basebenzise kanjani imidlalo yevidiyo ukuthuthukisa ucwaningo lwabo ku-AI.
Ungaphikisana ngokuthi, kusukela ekuqaleni kwayo, imidlalo yevidiyo ibe yindawo eshisayo yocwaningo kubacwaningi abanentshisekelo ku-AI.
Yize ingewona umdlalo wevidiyo osuka kuwo, i-chess ibigxile kakhulu ezinsukwini zokuqala ze-AI. Ngo-1951, uDkt. Dietrich Prinz wabhala uhlelo lokudlala i-chess esebenzisa ikhompyutha yedijithali ye-Ferranti Mark 1. Lokhu kwakusemuva esikhathini lapho la makhompuyutha amakhulu kufanele afunde izinhlelo ephepheni.
Uhlelo ngokwalo bekungelona i-chess AI ephelele. Ngenxa yokulinganiselwa kwekhompuyutha, i-Prinz ingakha kuphela uhlelo oluxazulula izinkinga ze-mate-in-two chess. Ngokwesilinganiso, uhlelo luthathe imizuzu eyi-15-20 ukubala konke ukunyakaza okungenzeka kwabadlali abaMhlophe nabaMnyama.
Sebenza ekuthuthukiseni i-chess namasheke I-AI ithuthuke kancane kuwo wonke amashumi eminyaka. Inqubekelaphambili yafinyelela umvuthwandaba ngo-1997 lapho i-IBM i-Deep Blue ihlula umkhulu we-chess waseRussia u-Garry Kasparov emdlalweni owodwa wemidlalo eyisithupha. Namuhla, izinjini ze-chess ongazithola kumakhalekhukhwini wakho zingahlula i-Deep Blue.
Abaphikisi be-AI baqale ukuthandwa ngesikhathi semidlalo ye-arcade yevidiyo. I-Space Invaders yango-1978 kanye ne-Pac-Man yawo-1980 amanye amavulandlela embonini ekudaleni i-AI engakwazi ukufaka inselelo ngokwanele ngisho nomakadebona wabadlali bama-arcade.
I-Pac-Man, ikakhulukazi, kwakuwumdlalo odumile wabacwaningi be-AI ababezozama ngawo. Okuhlukahlukene imincintiswano ngoba uNksz Pac-Man bahlelelwe ukucacisa ukuthi yiliphi iqembu elingaqhamuka ne-AI engcono kakhulu ukuhlula umdlalo.
I-Game AI nama-algorithms we-heuristic aqhubeka nokuvela njengoba kuphakama isidingo sabaphikisi abahlakaniphile. Isibonelo, ukulwa ne-AI kukhuphuke kakhulu ekudumeni njengoba izinhlobo ezifana nabadubuli bomuntu wokuqala ziba yinsakavukela.
Ukufunda Ngomshini Emidlalweni Yamavidiyo
Njengoba izindlela zokufunda zomshini zikhula ngokushesha ekudumeni, amaphrojekthi ahlukahlukene ocwaningo azame ukusebenzisa lawa masu amasha ukuze adlale imidlalo yevidiyo.
Imidlalo efana ne-Dota 2, i-StarCraft, ne-Doom ingasebenza njengezinkinga kulawa umshini wokufunda ama-algorithms ukuxazulula. Ama-algorithms wokufunda okujulile, ikakhulukazi, bakwazi ukufeza ngisho nokudlula ukusebenza kwezinga lomuntu.
The I-Arcade Learning Environment noma i-ALE inikeze abacwaningi isixhumi esibonakalayo semidlalo engaphezu kwekhulu ye-Atari 2600. Inkundla yomthombo ovulekile ivumele abacwaningi ukuthi balinganisele ukusebenza kwamasu okufunda komshini emidlalweni yevidiyo ye-Atari yakudala. I-Google ize yashicilela eyabo iphepha usebenzisa imidlalo eyisikhombisa evela ku-ALE
Phakathi naleso sikhathi, amaphrojekthi afana I-VizDoom inikeze abacwaningi be-AI ithuba lokuqeqesha ama-algorithms okufunda komshini ukuze badlale abadubuli bomuntu wokuqala be-3D.
Isebenza Kanjani: Eminye Imibono Ebalulekile
Amanethiwekhi eNeural
Izindlela eziningi zokuxazulula imidlalo yevidiyo ngokufunda komshini zibandakanya uhlobo lwe-algorithm eyaziwa ngokuthi inethiwekhi ye-neural.
Ungacabanga ngenethi ye-neural njengohlelo oluzama ukulingisa ukuthi ubuchopho bungase busebenze kanjani. Ngokufanayo nendlela ubuchopho bethu obakhiwe ngayo ama-neuron adlulisa isignali, inetha ye-neural nayo iqukethe ama-neuron okwenziwa.
Lawa ma-neuron okwenziwa aphinde adlulisele amasignali komunye nomunye, isiginali ngayinye iyinombolo yangempela. Inetha ye-neural iqukethe izendlalelo eziningi phakathi kwezingqimba zokufaka neziphumayo, ezibizwa ngokuthi inethiwekhi ye-neural ejulile.
Ukuqiniswa kokufunda
Enye indlela evamile yokufunda umshini ehambisana nokufunda imidlalo yevidiyo umbono wokuqinisa ukufunda.
Le nqubo iyinqubo yokuqeqesha i-ejenti kusetshenziswa imivuzo noma izijeziso. Ngale ndlela, i-ejenti kufanele ikwazi ukuqhamuka nesixazululo senkinga ngokuzama nangephutha.
Ake sithi sifuna i-AI ukuze sithole ukuthi idlalwa kanjani igeyimu yeNyoka. Inhloso yomdlalo ilula: thola amaphuzu amaningi ngokudla izinto futhi ugweme umsila wakho okhulayo.
Ngokufunda kokuqinisa, singachaza umsebenzi wokuklomelisa R. Umsebenzi wengeza amaphuzu lapho Inyoka idla into futhi ikhipha amaphuzu lapho Inyoka ishaya isithiyo. Uma kubhekwa indawo yamanje kanye nesethi yezenzo ezingenzeka, imodeli yethu yokufunda yokuqinisa izozama ukubala 'inqubomgomo' efanele eyenza umsebenzi wethu womvuzo ube mkhulu.
I-Neuroevolution
Ukugcina ingqikithi nokugqugquzelwa imvelo, abacwaningi bathole impumelelo ekusebenziseni i-ML emidlalweni yevidiyo ngokusebenzisa inqubo eyaziwa ngokuthi i-neuroevolution.
Esikhundleni sokusebenzisa ukwehla kwe-gradient ukuze sibuyekeze ama-neurons kunethiwekhi, singasebenzisa ama-algorithms wokuziphendukela kwemvelo ukuze sithole imiphumela engcono.
Ama-algorithms wokuziphendukela kwemvelo ngokuvamile aqala ngokukhiqiza inani labantu bokuqala abangahleliwe. Sibe sesihlola laba bantu sisebenzisa imibandela ethile. Abantu abangcono kakhulu bakhethwa “njengabazali” futhi bakhuliswa ndawonye ukuze bakhe isizukulwane esisha sabantu ngabanye. Laba bantu bazobe sebethatha isikhundla sabantu abalingana kancane emphakathini.
Lawa ma-algorithms ajwayele ukwethula uhlobo oluthile lomsebenzi wokuguqula ngesikhathi se-crossover noma isinyathelo "sokuzalela" ukuze kugcinwe ukuhlukahluka kofuzo.
Ucwaningo Lwesampula Lokufunda Ngomshini Kumageyimu Wevidiyo
I-OpenAI Eyisihlanu
I-OpenAI Eyisihlanu wuhlelo lwekhompyutha lwe-OpenAI oluhlose ukudlala i-DOTA 2, umdlalo odumile wenkundla yempi yeselula (MOBA).
Uhlelo lusebenzise izindlela zokufunda zokuqinisa ezikhona, ezenzelwe ukufunda ezigidini zozimele ngomzuzwana. Ngenxa yesistimu yokuqeqeshwa esabalalisiwe, i-OpenAI ikwazile ukudlala imidlalo yeminyaka engu-180 usuku ngalunye.
Ngemuva kwesikhathi sokuqeqeshwa, i-OpenAI Five yakwazi ukuzuza ukusebenza kwezinga lochwepheshe futhi yabonisa ukubambisana nabadlali abangabantu. Ngo-2019, i-OpenAI emihlanu yakwazi ukunqoba U-99.4% wabadlali emidlalweni yomphakathi.
Kungani i-OpenAI inqume ngalo mdlalo? Ngokusho kwabacwaningi, i-DOTA 2 yayinomshini oyinkimbinkimbi owawungaphandle kokufinyelela ekujuleni okukhona ukuqinisa ukufunda ubuchule obuphezulu.
Super Mario Bros.
Olunye uhlelo oluthokozisayo lwamanethi e-neural emidlalweni yevidiyo ukusetshenziswa kwe-neuroevolution ukudlala amapulatifomu afana neSuper Mario Bros.
Isibonelo, lokhu ukungena kwe-hackathon iqala ngokungabi nolwazi lomdlalo futhi yakha kancane kancane isisekelo salokho okudingekayo ukuze uthuthuke ngezinga.
I-neural net ezishintshayo ithatha isimo samanje segeyimu njengegridi yamathayela. Ekuqaleni, i-neural net ayiqondi ukuthi ithayela ngalinye lisho ukuthini, kuphela ukuthi amathayela “omoya” ahlukile “kumathayela aphansi” kanye “namathayela esitha.”
Ukuqaliswa kwephrojekthi ye-hackathon ye-neuroevolution kusebenzise i-algorithm yofuzo ye-NEAT ukuzalanisa amanethi emizwa ehlukene ngokukhetha.
Ukubaluleka
Manje njengoba usubone ezinye izibonelo zamanethi e-neural adlala imidlalo yevidiyo, ungahle uzibuze ukuthi liyini iphuzu lakho konke lokhu.
Njengoba imidlalo yevidiyo ibandakanya ukusebenzisana okuyinkimbinkimbi phakathi kwama-ejenti nendawo yabo, iyindawo yokuhlola ephelele yokwenza i-AI. Izindawo ezibonakalayo ziphephile futhi ziyalawuleka futhi zinikeza ukunikezwa kwedatha okungapheli.
Ucwaningo olwenziwe kulo mkhakha lunikeze abacwaningi ukuqonda kokuthi amanetha e-neural angenziwa kanjani kahle ukuze afunde ukuxazulula izinkinga emhlabeni wangempela.
Ama-Neural amanethiwekhi zigqugquzelwa indlela ubuchopho obusebenza ngayo emhlabeni wemvelo. Ngokufunda ukuthi ama-neurons okwenziwa aziphatha kanjani lapho efunda ukudlala umdlalo wevidiyo, singathola nokuqonda kokuthi ubuchopho bomuntu isebenza.
Isiphetho
Ukufana phakathi kwamanethiwekhi e-neural kanye nengqondo kuye kwaholela ekuqondeni kuyo yomibili le mikhakha. Ucwaningo oluqhubekayo lokuthi amanetha e-neural angaxazulula kanjani izinkinga ngolunye usuku lungaholela ezinhlotsheni ezithuthuke kakhulu ze ukuhlakanipha okungekhona okwangempela.
Cabanga usebenzisa i-AI ehambisana nokucaciswa kwakho engadlala wonke umdlalo wevidiyo ngaphambi kokuwuthenga ukuze ukwazise ukuthi ikufanele yini isikhathi sakho. Ingabe izinkampani zegeyimu yevidiyo zingasebenzisa amanethi e-neural ukuthuthukisa idizayini yegeyimu, ileveli ye-tweak, nobunzima bomphikisi?
Ucabanga ukuthi kuzokwenzekani uma amanetha e-neural eba abadlali bokugcina?
shiya impendulo