Jàngat buy dooleel bu dëddu
Jàngat buy dooleel bu dëddu (IRL) dafay soppi RL buñ miin: ludul ñu jox ko neexal ba noppi wut politik, dafay xool doxalinu eksper bi ba noppi di génne fonction neexal bu nëbbu bi koy leeral.
Résumé
This matters because a recovered reward generalizes to new situations far better than directly copied actions.
Plongeur bu xóot
Jàng-jàngat buy jàppale nit ñi dafay laaj: ban mébet la eksper bi wara topp ngir mëna doxalee ni ñu ko doxalee? Suñu joxee ay firnde, IRL dafay delloosi ab fonction recompense ci muy nuru lu baax (wala lu jege-gën), ba noppi jëfandikoo RL standard ngir génne ab politik. Motivation bi mooy generalisation - ab neexal buñu jàng mooy jàpp li waral ci ginaaw jeffin, suko defee agent bi mën jëfandikoo xel mu dal ci réew yi demonstration yi musul muur, wuute ak cloning behavioral biy toppandoo jëf. Jafe-jafe bi amul benn werante: fonction yu bari yuy neexal dañuy leeral benn jeffin, ba ci yu amul benn njariñ. Xeetu jëfandikoo bu am solo bi mooy saafara jafe-jafe yii, lu ci melni pexe yu gëna mag yu taamu neexal yi gëna leer ci eksper bi, ak IRL entropi bu gëna mag, biy tànn distribution neexal bi gëna néew lu méngoo ak done yi.
Gis-gis xarala
Jafe-jafe bi gëna mag mooy ñàkka leer: neexal bu amul benn njariñ dafay tax bépp politik gëna baax, kon neexal yu bari yu amul àpp ñooy leeral bépp wane. Entropie bu gëna mag IRL dafay saafara jafe-jafe yi ci modele wane yu bawoo ci distribution fu probabilite trajectoire di màgg exponentiellement ak neexal bu mat sëkk. Loolu dafay jur benn mébet bu amul fenn, bu leer te mën na jëflante ak kàngam yu bari xumbaay, ndax trajectoire suboptimal yi dañuy am probabilite bu gëna ndaw waaye du nul, moo gën ñu leen di dàq.
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 jàngat buy dooleel
IRL dafay gëna jàppale njàngum neexal ngir méngale: du nit ñi di kode neexal, sistem yi dañuy xam li nit ñi jox solo ci doxalin ak feedback. Xaarandi lëkkaloo yu gëna dëgër ak njàngum dooleel ci feedback nit ak njàngum taamu, escaleer ci modelu làkk ak jekkal robotik. Gëstu dafay puus ngir am neexal ci wideo yu ñor ak seetlu yu xaaj, ak ngir am neexal yuñ mëna xàmmee yuy baña nangu neexal-hacking ak jafe-jafe ambiguité yiy sonal tay xeetu neexal yi.
Doxal ci àdduna dëgg
Oto yuy demal seen bopp di jël tànneefi dawal (nooy, marse kaaraange) ci dawalkat nit
Robot yi dañuy jàng mébetu liggéey ci ni nit ñi di wane ngir yamale ci jëmmal yu bees
Modeling doxantukat yi wala baayima yi ci delloosi mébet yi ci ginaaw trajectoire yiñ seetlu
Neexal inference ngir IA alignment, jàng valëri doomi aadama ci tanneef yuñ wane
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 Inverse Reinforcement Learning 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
Jàngat bu am doole ci feedback nit
Laaj yi ñuy faral di laaj
What is Inverse Reinforcement Learning?
Jàngat buy dooleel bu dëddu (IRL) dafay soppi RL buñ miin: ludul ñu jox ko neexal ba noppi wut politik, dafay xool doxalinu eksper bi ba noppi di génne fonction neexal bu nëbbu bi koy leeral. Loolu dafa am solo ndax neexal biñ delloo dafay gëna jëm ci jafe-jafe yu bees, moo gën jëf yuñ kopie ci saasi.
Lan mooy njàngum dooleel bu dëddu ngir delloosi?
IRL dafay jël firnde yi ni ay done ba noppi di tënk fonction neexal bi ci suufu doxalinu eksper bi di nuru lu gëna baax.
Lan moo waral ñuy wax ni jafe-jafe IRL du lu baax wala lu jaxasoo?
Fonction neexal yu bari, yu ci melni neexal bu amul benn njariñ, mën na tax jeffin ju ñuy seetlu gëna baax, kon neexal bi du nekk luy wane rek.
Lan mooy njariñu delloosi neexal ci klonaasu jeffin?
Benn fonction neexal dafay kode li waral expert bi def limu def, bàyyi politik biñ jëlee mu doxalee ci anam wu jaar yoon ci etaa yu bees yi, waaye klonaas bi dafay kopie jëf yiñ gis ci done yi.
naka lay def ba IRL entropi bu gëna mag di saafara jafe-jafey neexal?
Max-entropy IRL dafay xalaat ni trajectoire yu gëna am neexal ñoo gëna am wërsëg, di joxe mébet bu amul fenn buy muñ eksper yu bari te bari xumbaay.
Ginaaw IRL delloosi ab fonction neexal, lan lañuy faral di def ci topp?
IRL dafay defar neexal, ñu jox ko algorithm RL bu jaar yoon ngir xayma politik bu baax ci wàllu mébet biñ jël.