DeepMind, wanda aka sani don ci gaba a fagen ilimin wucin gadi da gina hanyoyin sadarwa na jijiyoyi masu iya yin wasan kwamfuta da wasan allo a matakin ɗan adam, kwanan nan an bayyana aikin AlphaCode wanda ya bayyana yadda tsarin koyon injin don tsara code cewa zaku iya shiga cikin gasa na shirye-shirye akan dandamali na Codeforces kuma ku nuna matsakaicin sakamako.
An ambaci cewa aikin yana amfani da tsarin gine-ginen cibiyar sadarwa na "Transformer". a haɗe tare da wasu hanyoyin samfura da tacewa don samar da bambance-bambancen lambobi daban-daban waɗanda ba za su iya faɗi ba daidai da rubutun yare na halitta.
Hanyar yadda yake aiki AlphaCode ta dogara ne a kan tacewa, haɗawa da rarrabuwa, sannan ta ci gaba da zaɓar mafi kyawun lambar aiki daga rafi na zaɓuɓɓukan da aka samar, sannan a bincika don tabbatar da cewa an sami sakamako mai kyau (a cikin kowane aiki na gasar, misali. bayanan shigarwa da sakamako mai dacewa) zuwa wannan misali, wanda ya kamata a samu bayan aiwatar da shirin).
Muna dalla-dalla AlphaCode, wanda ke amfani da ƙirar yare na tushen canji don samar da lamba a sikelin da ba a taɓa ganin irinsa ba, sannan a hankali yana tace ƙaramin tsari na shirye-shirye masu ban sha'awa.
Muna inganta ayyukanmu ta amfani da gasa da aka shirya akan Codeforces, sanannen dandamali wanda ke ɗaukar nauyin gasa na yau da kullun waɗanda ke jan hankalin dubun dubatar masu shigowa daga ko'ina cikin duniya waɗanda ke zuwa don gwada ƙwarewar coding. Mun zaɓi gasa 10 kwanan nan don kimantawa, kowane sabo fiye da bayanan horonmu. AlphaCode ya kasance kusan matakin tare da matsakaita mai fafatawa, alama ce ta farko da tsarin tsara lambar AI ya kai matakin gasa a gasar shirye-shirye.
Domin horon tsarin kusan ilimin injin, an haskaka cewa an yi amfani da lambar tushe da ke cikin wuraren ajiyar jama'a na GitHub. Bayan shirya samfurin farko, an aiwatar da tsarin ingantawa bisa tarin lambar tare da misalan matsaloli da mafita da aka ba wa mahalarta gasar Codeforces, CodeChef, HackerEarth, AtCoder da Aizu.
Gabaɗaya, don ƙirƙirar AlphaCode 715 GB na lambar GitHub da aka yi amfani da shi da misalan sama da miliyan guda na hanyoyin magance matsalolin da suka shafi gasar. Kafin a ci gaba zuwa tsara tsara lambar, rubutun aikin ya shiga cikin wani lokaci na daidaitawa, wanda aka cire duk abin da ya wuce gona da iri kuma kawai manyan sassa sun rage.
Don gwada tsarin, an zaɓi sababbin 10 sababbin Codeforces tare da mahalarta fiye da 5.000, wanda aka gudanar bayan kammala horar da samfurin koyon inji.
Zan iya faɗi a amince cewa sakamakon AlphaCode ya wuce tsammanina. Na yi shakka saboda ko da a cikin matsalolin gasa masu sauƙi, ana buƙatar sau da yawa ba kawai don aiwatar da algorithm ba, amma kuma (kuma wannan shine mafi wuya) don ƙirƙira shi. AlphaCode ya sami nasarar yin aiki a matakin sabon mai fafatawa. Ba zan iya jira in ga abin da zai zo ba!
MIKE MIRZAYANOV
Wanda ya kafa CODEFORCES
Sakamakon ayyukan da aka ba da izini don tsarin AlphaCode ya shiga kusan a tsakiyar cancantar wadannan gasa (54,3%). AlfaCode ta annabta gabaɗayan maki shine maki 1238, yana ba da tabbacin shiga cikin Manyan 28% a tsakanin duk mahalarta Codeforces waɗanda suka halarci gasa aƙalla sau ɗaya a cikin watanni 6 da suka gabata.
Ya kamata a lura cewa an lura cewa aikin har yanzu yana cikin mataki na farko na ci gaba da kuma cewa a nan gaba an tsara shi don inganta ingancin lambar da aka samar, da kuma inganta AlphaCode zuwa tsarin da ke taimakawa wajen rubuta lambar. ko kayan aikin haɓaka aikace-aikace waɗanda mutanen da ba su da ƙwarewar shirye-shirye za su iya amfani da su.
Finalmente idan kuna sha'awar ƙarin sani game da shi, Ya kamata ku sani cewa muhimmin fasalin haɓakawa shine ikon samar da code a Python ko C++, ɗaukar matsayin shigar da rubutu bayanin matsalar cikin Ingilishi.
Kuna iya duba cikakkun bayanai A cikin mahaɗin mai zuwa.