No-regret exploration in goal-oriented ...
Document type :
Communication dans un congrès avec actes
Title :
No-regret exploration in goal-oriented reinforcement learning
Author(s) :
Tarbouriech, Jean [Auteur]
Scool [Scool]
Facebook AI Research [Paris] [FAIR]
Garcelon, Evrard [Auteur]
Facebook AI Research [Paris] [FAIR]
Valko, Michal [Auteur]
DeepMind [Paris]
Pirotta, Matteo [Auteur]
Facebook AI Research [Paris] [FAIR]
Lazaric, Alessandro [Auteur]
Facebook AI Research [Paris] [FAIR]
Scool [Scool]
Facebook AI Research [Paris] [FAIR]
Garcelon, Evrard [Auteur]
Facebook AI Research [Paris] [FAIR]
Valko, Michal [Auteur]
DeepMind [Paris]
Pirotta, Matteo [Auteur]
Facebook AI Research [Paris] [FAIR]
Lazaric, Alessandro [Auteur]
Facebook AI Research [Paris] [FAIR]
Conference title :
International Conference on Machine Learning
City :
Vienna / Virtual
Country :
Autriche
Start date of the conference :
2020
HAL domain(s) :
Statistiques [stat]/Machine Learning [stat.ML]
English abstract : [en]
Many popular reinforcement learning problems (e.g., navigation in a maze, some Atari games, mountain car) are instances of the episodic setting under its stochastic shortest path (SSP) formulation, where an agent has to ...
Show more >Many popular reinforcement learning problems (e.g., navigation in a maze, some Atari games, mountain car) are instances of the episodic setting under its stochastic shortest path (SSP) formulation, where an agent has to achieve a goal state while minimizing the cumulative cost. Despite the popularity of this setting, the explorationexploitation dilemma has been sparsely studied in general SSP problems, with most of the theoretical literature focusing on different problems (i.e., finite-horizon and infinite-horizon) or making the restrictive loop-free SSP assumption (i.e., no state can be visited twice during an episode). In this paper, we study the general SSP problem with no assumption on its dynamics (some policies may actually never reach the goal). We introduce UC-SSP, the first no-regret algorithm in this setting, and prove a regret bound scaling as O(DS √ ADK) after K episodes for any unknown SSP with S states, A actions, positive costs and SSP-diameter D, defined as the smallest expected hitting time from any starting state to the goal. We achieve this result by crafting a novel stopping rule, such that UC-SSP may interrupt the current policy if it is taking too long to achieve the goal and switch to alternative policies that are designed to rapidly terminate the episode.Show less >
Show more >Many popular reinforcement learning problems (e.g., navigation in a maze, some Atari games, mountain car) are instances of the episodic setting under its stochastic shortest path (SSP) formulation, where an agent has to achieve a goal state while minimizing the cumulative cost. Despite the popularity of this setting, the explorationexploitation dilemma has been sparsely studied in general SSP problems, with most of the theoretical literature focusing on different problems (i.e., finite-horizon and infinite-horizon) or making the restrictive loop-free SSP assumption (i.e., no state can be visited twice during an episode). In this paper, we study the general SSP problem with no assumption on its dynamics (some policies may actually never reach the goal). We introduce UC-SSP, the first no-regret algorithm in this setting, and prove a regret bound scaling as O(DS √ ADK) after K episodes for any unknown SSP with S states, A actions, positive costs and SSP-diameter D, defined as the smallest expected hitting time from any starting state to the goal. We achieve this result by crafting a novel stopping rule, such that UC-SSP may interrupt the current policy if it is taking too long to achieve the goal and switch to alternative policies that are designed to rapidly terminate the episode.Show less >
Language :
Anglais
Peer reviewed article :
Oui
Audience :
Internationale
Popular science :
Non
Collections :
Source :
Files
- https://hal.inria.fr/hal-03287824/document
- Open access
- Access the document
- https://hal.inria.fr/hal-03287824/document
- Open access
- Access the document
- https://hal.inria.fr/hal-03287824/document
- Open access
- Access the document
- document
- Open access
- Access the document
- tarbouriech2020no-regret.pdf
- Open access
- Access the document