M'ndandanda wazopezekamo[Bisani][Show]
Tiyerekeze kuti mukuyesera kuphunzitsa loboti kuyenda. Mosiyana ndi kuphunzitsa makompyuta mmene angadziwiretu mitengo ya katundu kapena kugawa zithunzi m’magulu, tilibe deta yaikulu imene tingagwiritse ntchito pophunzitsa loboti yathu.
Ngakhale kuti zingabwere mwachibadwa kwa inu, kuyenda ndi chinthu chovuta kwambiri. Kuyenda sitepe kumaphatikizapo minofu yambirimbiri yogwirira ntchito limodzi. Khama ndi njira zomwe zimagwiritsidwa ntchito poyenda kuchokera kumalo ena kupita kwina zimadaliranso zinthu zosiyanasiyana, kuphatikizapo ngati mwanyamula chinachake kapena pali chopinga kapena mitundu ina ya zopinga.
Muzochitika ngati izi, titha kugwiritsa ntchito njira yotchedwa reinforcement learning kapena RL. Ndi RL, mukhoza kufotokozera cholinga chenicheni chomwe mukufuna kuti chitsanzo chanu chithetse ndipo pang'onopang'ono mulole chitsanzocho chiphunzire pachokha momwe mungakwaniritsire.
M'nkhaniyi, tiwona zoyambira zolimbikitsira kuphunzira komanso momwe tingagwiritsire ntchito dongosolo la RL pamavuto osiyanasiyana padziko lapansi.
Kodi maphunziro owonjezera ndi chiyani?
Maphunziro olimbikitsa amatanthauza kagawo kakang'ono ka makina kuphunzira zomwe zimayang'ana kwambiri pakupeza mayankho mwa kudalitsa machitidwe omwe mukufuna komanso kulanga machitidwe osayenera.
Mosiyana ndi kuphunzira koyang'aniridwa, njira yophunzirira yolimbikitsira nthawi zambiri ilibe gulu lophunzitsira lomwe limapereka zotulukapo zoyenera pazolowera zomwe zaperekedwa. Popanda chidziwitso cha maphunziro, algorithm iyenera kupeza yankho mwa kuyesa ndi zolakwika. Algorithm, yomwe nthawi zambiri timayitcha kuti wothandizila, ayenera kupeza yankho palokha mwa kucheza ndi environment.
Ochita kafukufuku amasankha zomwe zidzachitike mphotho ndi zomwe algorithm imatha kuchita. Aliyense kuchitapo ma aligorivimu amatenga adzalandira mtundu wina wa mayankho omwe amawonetsa momwe ma aligorivimu akuchitira. Panthawi yophunzitsira, algorithm pamapeto pake idzapeza njira yabwino yothetsera vuto linalake.
Chitsanzo Chosavuta: 4 × 4 Gridi
Tiyeni tiwone chitsanzo chosavuta cha vuto lomwe tingathe kuthana ndi maphunziro olimbikitsa.
Tiyerekeze kuti tili ndi gridi ya 4 × 4 monga malo athu. Wothandizira wathu amayikidwa mwachisawawa mu umodzi mwamabwalo pamodzi ndi zopinga zingapo. Gululi liyenera kukhala ndi zopinga zitatu za "dzenje" zomwe ziyenera kupewedwa ndi mphotho imodzi ya "diamondi" yomwe wothandizira ayenera kupeza. Kufotokozera kwathunthu kwa chilengedwe chathu kumadziwika kuti chilengedwe Boma.
Muchitsanzo chathu cha RL, wothandizira wathu amatha kupita kumalo aliwonse oyandikana nawo malinga ngati palibe zopinga zomwe zimawalepheretsa. Seti ya zochita zonse zovomerezeka m'malo omwe mwapatsidwa amadziwika kuti ndi malo ochitirapo kanthu. Cholinga cha wothandizira wathu ndikupeza njira yaifupi kwambiri yopita ku mphotho.
Wothandizira wathu adzagwiritsa ntchito njira yolimbikitsira yophunzirira kuti apeze njira yopita ku diamondi yomwe imafuna masitepe ochepa. Njira iliyonse yoyenera idzapatsa robot mphotho ndipo sitepe iliyonse yolakwika idzachotsa mphotho ya robot. Chitsanzocho chimawerengera malipiro onse pamene wothandizira afika ku diamondi.
Tsopano popeza tafotokozera wothandizila ndi chilengedwe, tiyeneranso kufotokozera malamulo oti tigwiritse ntchito podziwa zomwe wothandizi angachite potengera momwe alili komanso chilengedwe.
Ndondomeko ndi Mphotho
Mu chitsanzo cholimbikitsa maphunziro, a policy kutanthauza njira yomwe wothandizila amagwiritsa ntchito kuti akwaniritse zolinga zawo. Ndondomeko ya wothandizirayo ndi yomwe imasankha zomwe wothandizira ayenera kuchita potengera momwe wodwalayo alili komanso chilengedwe chake.
Wothandizira ayenera kuwunika ndondomeko zonse zomwe zingatheke kuti awone ndondomeko yomwe ili yabwino.
Muchitsanzo chathu chophweka, kutera pamalo opanda kanthu kudzabweza mtengo wa -1. Wothandizirayo akafika pamalo omwe ali ndi mphotho ya diamondi, adzalandira mtengo wa 10. Pogwiritsa ntchito mfundozi, tikhoza kuyerekezera ndondomeko zosiyanasiyana pogwiritsa ntchito ntchito zothandiza U.
Tiyeni tifananize kugwiritsa ntchito mfundo ziwiri zomwe tawona pamwambapa:
U(A) = -1 – 1 -1 + 10 = 7
U(B) = -1 – 1 – 1 – 1 – 1 + 10 = 5
Zotsatira zikuwonetsa kuti Policy A ndiyo njira yabwinoko yopezera mphotho. Chifukwa chake, wothandizira adzagwiritsa ntchito Njira A kuposa Policy B.
Kufufuza motsutsana ndi Kugwiritsa Ntchito Masuku pamutu
Vuto lofufuza ndi kugwiritsa ntchito masuku pamutu pakulimbikitsa kuphunzira ndi vuto lomwe wothandizira ayenera kukumana nalo panthawi yopanga chisankho.
Kodi othandizira ayenera kuyang'ana njira zatsopano kapena zosankha kapena apitilize kugwiritsa ntchito zomwe akudziwa kale?
Ngati wothandizira asankha kufufuza, pali kuthekera kwa wothandizira kupeza njira yabwinoko, koma zikhoza kuwononganso nthawi ndi chuma. Kumbali ina, ngati wothandizira asankha kugwiritsa ntchito yankho lomwe akudziwa kale, akhoza kuphonya njira yabwinoko.
Mapulogalamu Othandiza
Nazi njira zina Ofufuza a AI agwiritsa ntchito zitsanzo zolimbikitsira kuti athetse mavuto adziko lenileni:
Kulimbikitsa Kuphunzira M'magalimoto Odziyendetsa Okha
Maphunziro olimbikitsa agwiritsidwa ntchito pa magalimoto odziyendetsa okha kuti athe kuyendetsa bwino ndi kuyendetsa bwino. Ukadaulo umathandizira magalimoto odziyimira pawokha kuphunzira kuchokera ku zolakwa zawo ndikusintha mosalekeza machitidwe awo kuti akwaniritse bwino ntchito yawo.
Mwachitsanzo, kampani ya AI yochokera ku London Wayve wagwiritsa ntchito bwino njira yophunzirira yolimbikitsira yoyendetsa galimoto. Pakuyesa kwawo, adagwiritsa ntchito mphotho yomwe imakulitsa kuchuluka kwa nthawi yomwe galimoto imathamanga popanda woyendetsa kuti aperekepo.
Mitundu ya RL imathandizanso magalimoto kupanga zisankho motengera chilengedwe, monga kupewa zopinga kapena kuphatikiza magalimoto. Zitsanzozi ziyenera kupeza njira yosinthira malo ovuta ozungulira galimoto kukhala malo oimira boma omwe chitsanzocho chingamvetse.
Kulimbitsa Maphunziro mu Robotics
Ofufuza akhala akugwiritsanso ntchito maphunziro olimbikitsa kupanga maloboti omwe amatha kuphunzira ntchito zovuta. Kupyolera mu zitsanzo za RL izi, maloboti amatha kuyang'ana chilengedwe chawo ndikupanga zisankho potengera zomwe akuwona.
Mwachitsanzo, kafukufuku wapangidwa pakugwiritsa ntchito njira zolimbikitsira kuphunzira kulola maloboti a bipedal kuphunzira momwe angachitire kuyenda paokha.
Ochita kafukufuku amawona kuti RL ndi njira yofunikira kwambiri pankhani ya robotics. Kuphunzira kolimbikitsa kumapatsa othandizira ma robot dongosolo kuti aphunzire zinthu zapamwamba zomwe zingakhale zovuta kupanga mainjiniya.
Kulimbitsa Maphunziro mu Masewera
Mitundu ya RL yagwiritsidwanso ntchito pophunzira kusewera masewera apakanema. Ma Agents atha kukhazikitsidwa kuti aphunzire kuchokera ku zolakwa zawo ndikusintha mosalekeza machitidwe awo pamasewera.
Ofufuza apanga kale zida zomwe zimatha kusewera masewera monga chess, Go, ndi poker. Mu 2013, Kuzama anagwiritsa ntchito Deep Reinforcement Learning kulola chitsanzo kuti aphunzire kusewera masewera a Atari kuyambira pachiyambi.
Masewera ambiri a board ndi masewera apakanema ali ndi malo ochepa ochitirapo kanthu komanso cholinga chodziwika bwino cha konkriti. Makhalidwewa amagwira ntchito ku ubwino wa chitsanzo cha RL. Njira za RL zimatha kubwereza mwachangu masewera oyerekeza mamiliyoni ambiri kuti aphunzire njira zabwino zopezera chipambano.
Kutsiliza
Kaya ndikuphunzira kuyenda kapena kuphunzira kusewera masewera apakanema, zitsanzo za RL zatsimikiziridwa kuti ndizothandiza za AI zothetsa mavuto omwe amafunikira kupanga zisankho zovuta.
Pamene teknoloji ikupitirizabe kusinthika, onse ofufuza ndi opanga mapulogalamu apitirizabe kupeza ntchito zatsopano zomwe zimatengera luso lachitsanzo kudziphunzitsa.
Ndi ntchito ziti zothandiza zomwe mukuganiza kuti maphunziro olimbikitsira angathandize?
Siyani Mumakonda