Matches in SemOpenAlex for { <https://semopenalex.org/work/W3008082783> ?p ?o ?g. }
- W3008082783 abstract "Bayesian reward learning from demonstrations enables rigorous safety and uncertainty analysis when performing imitation learning. However, Bayesian reward learning methods are typically computationally intractable for complex control problems. We propose Bayesian Reward Extrapolation (Bayesian REX), a highly efficient Bayesian reward learning algorithm that scales to high-dimensional imitation learning problems by pre-training a low-dimensional feature encoding via self-supervised tasks and then leveraging preferences over demonstrations to perform fast Bayesian inference. Bayesian REX can learn to play Atari games from demonstrations, without access to the game score and can generate 100,000 samples from the posterior over reward functions in only 5 minutes on a personal laptop. Bayesian REX also results in imitation learning performance that is competitive with or better than state-of-the-art methods that only learn point estimates of the reward function. Finally, Bayesian REX enables efficient high-confidence policy evaluation without having access to samples of the reward function. These high-confidence performance bounds can be used to rank the performance and risk of a variety of evaluation policies and provide a way to detect reward hacking behaviors." @default.
- W3008082783 created "2020-03-06" @default.
- W3008082783 creator A5042359413 @default.
- W3008082783 creator A5043572737 @default.
- W3008082783 creator A5057137533 @default.
- W3008082783 creator A5077785918 @default.
- W3008082783 date "2020-02-21" @default.
- W3008082783 modified "2023-09-27" @default.
- W3008082783 title "Safe Imitation Learning via Fast Bayesian Reward Inference from Preferences" @default.
- W3008082783 cites W1522301498 @default.
- W3008082783 cites W1591675293 @default.
- W3008082783 cites W1777239053 @default.
- W3008082783 cites W1845972764 @default.
- W3008082783 cites W1986014385 @default.
- W3008082783 cites W1999874108 @default.
- W3008082783 cites W2051228319 @default.
- W3008082783 cites W2061562262 @default.
- W3008082783 cites W2097381042 @default.
- W3008082783 cites W2098774185 @default.
- W3008082783 cites W2103850933 @default.
- W3008082783 cites W2111051539 @default.
- W3008082783 cites W2116671302 @default.
- W3008082783 cites W2145339207 @default.
- W3008082783 cites W2155027007 @default.
- W3008082783 cites W2167696704 @default.
- W3008082783 cites W2168024904 @default.
- W3008082783 cites W2273088453 @default.
- W3008082783 cites W2410842990 @default.
- W3008082783 cites W2462906003 @default.
- W3008082783 cites W2467604901 @default.
- W3008082783 cites W2472245206 @default.
- W3008082783 cites W2566467060 @default.
- W3008082783 cites W2595845486 @default.
- W3008082783 cites W2600383743 @default.
- W3008082783 cites W2604173613 @default.
- W3008082783 cites W2734346390 @default.
- W3008082783 cites W2735318784 @default.
- W3008082783 cites W2736601468 @default.
- W3008082783 cites W2763110165 @default.
- W3008082783 cites W2768908787 @default.
- W3008082783 cites W2774527530 @default.
- W3008082783 cites W2798750840 @default.
- W3008082783 cites W2804066752 @default.
- W3008082783 cites W2807776404 @default.
- W3008082783 cites W2889990052 @default.
- W3008082783 cites W2900626361 @default.
- W3008082783 cites W2907574724 @default.
- W3008082783 cites W2912168444 @default.
- W3008082783 cites W2916286980 @default.
- W3008082783 cites W2945995507 @default.
- W3008082783 cites W2949115740 @default.
- W3008082783 cites W2949496227 @default.
- W3008082783 cites W2949933854 @default.
- W3008082783 cites W2962715211 @default.
- W3008082783 cites W2962717849 @default.
- W3008082783 cites W2962787969 @default.
- W3008082783 cites W2962799618 @default.
- W3008082783 cites W2962937519 @default.
- W3008082783 cites W2962943921 @default.
- W3008082783 cites W2962957031 @default.
- W3008082783 cites W2963099438 @default.
- W3008082783 cites W2963208223 @default.
- W3008082783 cites W2963238274 @default.
- W3008082783 cites W2963277051 @default.
- W3008082783 cites W2963289505 @default.
- W3008082783 cites W2963508354 @default.
- W3008082783 cites W2963590100 @default.
- W3008082783 cites W2963646405 @default.
- W3008082783 cites W2963836326 @default.
- W3008082783 cites W2963956018 @default.
- W3008082783 cites W2964055673 @default.
- W3008082783 cites W2964059111 @default.
- W3008082783 cites W2964094335 @default.
- W3008082783 cites W2964177756 @default.
- W3008082783 cites W2964263543 @default.
- W3008082783 cites W2964291307 @default.
- W3008082783 cites W2971130081 @default.
- W3008082783 cites W2980006775 @default.
- W3008082783 cites W2988072049 @default.
- W3008082783 cites W2989897153 @default.
- W3008082783 cites W3003342008 @default.
- W3008082783 cites W3022429412 @default.
- W3008082783 cites W3029228949 @default.
- W3008082783 cites W3036996956 @default.
- W3008082783 cites W2886585480 @default.
- W3008082783 hasPublicationYear "2020" @default.
- W3008082783 type Work @default.
- W3008082783 sameAs 3008082783 @default.
- W3008082783 citedByCount "17" @default.
- W3008082783 countsByYear W30080827832019 @default.
- W3008082783 countsByYear W30080827832020 @default.
- W3008082783 countsByYear W30080827832021 @default.
- W3008082783 crossrefType "posted-content" @default.
- W3008082783 hasAuthorship W3008082783A5042359413 @default.
- W3008082783 hasAuthorship W3008082783A5043572737 @default.
- W3008082783 hasAuthorship W3008082783A5057137533 @default.
- W3008082783 hasAuthorship W3008082783A5077785918 @default.
- W3008082783 hasConcept C105795698 @default.
- W3008082783 hasConcept C107673813 @default.
- W3008082783 hasConcept C119857082 @default.