Xeetu aktër-Critik
Pexem Actor-Critic dafay boole ñaari jàngat: 'actor' biy tànn jëf yi ak 'critique' biy àtte ni jëf yooyu baaxoon.
Résumé
This pairing makes reinforcement learning more stable and sample-efficient than using either approach alone.
Plongeur bu xóot
Njàngalem dooleel amna ñaari anam yu yaatu: njàngale yu lalu ci politik yuy jàngale ci saasi li ñu wara def, ak njàngale yu lalu ci valeur yuy jàngale ni réew yi baaxee. Aktër-Critique dañu leen boole. Aktër bi politik la buy génne probabilite jëf; critique bi fonction valeur la buy xayma ndàmpaay bi ñuy seentu. Ginaaw jéego bu nekk, ki koy ŋàññi dafay xayma njuumte ci diggante jamono ngir wane ndax njariñ li dafa baax wala moo gëna bon li ñu seentuwoon. Aktër bi dafay jëfandikoo njuumte boobu ngir puus politikam ci jëf yu raw li ñuy seentu, ba noppi sori ci jëf yu baaxul yi. Ndax te ŋàññikat bi dafa joxe ab baseline bu woyof, xayma gradient aktër bi moo gëna néew bruit ni ci pexe gradient politik yu seer yu melni REINFORCE, ci noonu lañuy jëfandikoo barabi jëf yuy wéy yu pexe yu am solo yu melni Q-Learning di gis ni dañu jaxasoo.
Gis-gis xarala
Aktër bi dafay yeesal ay parametram ci wàllu politik, ñu yokk ko ci njariñ A(s,a) = Q(s,a) - V(s), bi ŋàññikat bi xayma (dafay faral di jaar ci njuumte TD r + gamma*V(s) - V(s)). Njariñ li mooy wane ni jëf ji gënee baax ci li réew mi def, kon njariñ yu baax yi dañuy dooleel jëf yi, yu baaxul yi dañu leen di tere. Critique bi deñukoy tàggat boppam ngir wàññi njuumte TD bi.
njeextalu pexe
Njëgg ak budget
Dogal yi architecture di jël dañuy indi njariñ ak njëgu liggéey bi ay at ci ginaaw.
dogal yu gëna leer
Njàngalem xarala yi dafay jàppale ekip yi ñu tànn li gën, te baña yam ci li gëna bees daal.
Xool kalite
Tanneef yu gëna baax ci wàllu ingeñër dina wàññi jafe-jafe yi ci wàllu wóor ci liggéey bi.
Ëlëgu njuréefi aktër-Critique
Aktër-Critic mooy yax gi gëna am solo ci RL yu bees yi. Algorithm yu melni A3C, A2C, PPO, SAC, ak DDPG ñoom ñépp a ngi tabax ci kaw, yokk ay pexe yu melni mébet yuñ dagg ngir yeesali stabil, bonus entropi ngir banneexu, ak aktër yu paralel ngir génne. Xaarandil yokkute bu wéy ci robotik, agent jeu yu mag, ak RL ci feedback nit ñi ngir model làkk yi, fu stabilite ak sample efficacité gëna am solo.
Doxal ci àdduna dëgg
Taggat loxo robot ak njiitu locomotion ak couple yuy wéy di boole (lu melni, jëfandikoo PPO wala SAC)
Liggéeyu xeetu làkk yu yaatu jaaraleko ci RLHF, fu PPO (xeetu aktër-gëstu) di gëna baaxal tontu ci xeetu neexal
Am pomi pexe yu jaffe yu melni StarCraft II ak Dota 2
Data-centre seddal ak doxal energie di jàng ajustement yuy wéy
Risk yi ak balustrade yi
Optimize benn benchmark mën na nëbb ñakk kattan yu gëna yaatu ci sistem bi.
Njëg li ñuy fay ci infrastructure yi ak ci toppatoo dañuy faral di suufeel.
Bu sistem yi di gëna xawa jafee xam, jafe-jafe yi am ci wàllu kaaraange ak seetlu mën nañu gëna bari.
Roadmap ngir samp gi
Mandargal latency, kalite, ak njëg yi laata ngay jëfandikoo.
Benchmark ci biir sargal ak done yu dëggu.
Jumtukaay bi di saytu njuumte yi, derive bi ak njeextalu jëfandikukat bi.
Waajal rollback ak yooni tontu ci jafe-jafe yi laata ngay eskale.
Weyal di banneexu
Free newsletter
Get the daily AI briefing
Three verified AI stories every weekday morning, written in plain English. Free forever, no ads.
One email each weekday. Unsubscribe in one click. We never sell or share your address.
Test yourself
Take the Actor-Critic Methods quiz
Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.
Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation
Gis bi ci topp
Ñaareelu jaar-jaar bi gëna xéewale ak njuréefi Newton
Laaj yi ñuy faral di laaj
What is Actor-Critic Methods?
Pexem Actor-Critic dafay boole ñaari jàngat: 'actor' biy tànn jëf yi ak 'critique' biy àtte ni jëf yooyu baaxoon. Njaxas moomu dafay tax njàngum dooleel gëna dëgër, gëna am njariñ, moo gën jëfandikoo benn ci ñaari xeeti njàngale yooyu kese.
Ci anamu Actor-Critic, lan mooy wareefu 'critique' bi?
Critique bi dafay jàng benn fonction valeur ba noppi génne siñaal buy jàngat (lu melni njuumte TD) buy wax aktër bi ndax jëfam yi ñoo gënoon baax wala ñoo gënoon bon li ñu seentuwoon.
Lan mooy 'njariñ' A(s,a) = Q(s,a) - V(s)?
Njariñ li mooy xàjjale ba ñaata la benn jëf di gëna am njariñ ci njariñu réew mi, di joxe siñaal bu gëna set te baña dellu ginaaw.
Lan moo waral yokk benn ŋàññikat di wàññi faraas bi buñu ko méngale ak pexe yu sell yu melni REINFORCE?
Dindi valeur estimation critique bi muy ligne de base dafay wàññi variance estimation gradient bi te doo yokk benn parçage, di gaawlu jàng bi.
Ban kàttan la pexe Actor-Critic mëna am ci pexe yu lalu ci valeur yu melni Q-Learning ñu mëna jëfandikoo ci anam wu jaxasoo?
Ndax aktër bi dafay parametre ci saasi benn politik, mën na génne jëf yu wéy (lu melni, couple yuñ boole) te soxlawul max ci jëf yu bari te amul àpp.
Ban siñaal la aktër bi di faral di jëfandikoo ngir yeesal ay sàrtam?
Aktër bi dafay yamale politikam ci wàll wi siñaal njariñu ŋàññikat bi/TD-njuumte digle, di dooleel jëf yu gëna baax yuñu doon seentu.