An official website of the United States government
Comprehensive state-action exploration is essential for reinforcement learning (RL) algorithms. It enables them to find optimal solutions and avoid premature convergence. In value-based RL, optimistic initialization of the value function ensures sufficient exploration for finding the optimal solution. Optimistic values lead to curiosity-driven exploration enabling visitation of under-explored regions. However, optimistic initialization has limitations in stochastic and non-stationary environments due to its inability to explore ''infinitely-often''. To address this limitation, we propose a novel exploration strategy for value-based RL, denoted COIN, based on recurring optimistic initialization. By injecting a continual exploration bonus, we overcome the shortcoming of optimistic initialization (sensitivity to environment noise). We provide a rigorous theoretical comparison of COIN versus existing popular exploration strategies and prove it provides a unique set of attributes (coverage, infinite-often, no visitation tracking, and curiosity). We demonstrate the superiority of COIN over popular existing strategies on a designed toy domain as well as present results on common benchmark tasks. We observe that COIN outperforms existing exploration strategies in four out of six benchmark tasks while performing on par with the best baseline on the other two tasks.
more »
« less
Optimistic Initialization for Exploration in Continuous Control
Optimistic initialization underpins many theoretically sound exploration schemes in tabular domains; however, in the deep function approximation setting, optimism can quickly disappear if initialized naively. We propose a framework for more effectively incorporating optimistic initialization into reinforcement learning for continuous control. Our approach uses metric information about the state-action space to estimate which transitions are still unexplored, and explicitly maintains the initial Q-value optimism for the corresponding state-action pairs. We also develop methods for efficiently approximating these training objectives, and for incorporating domain knowledge into the optimistic envelope to improve sample efficiency. We empirically evaluate these approaches on a variety of hard exploration problems in continuous control, where our method outperforms existing exploration techniques.
more »
« less
- PAR ID:
- 10321086
- Date Published:
- Journal Name:
- Proceedings of the Thirty-Sixth AAAI Conference on Artificial Intelligence
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Safe reinforcement learning is extremely challenging--not only must the agent explore an unknown environment, it must do so while ensuring no safety constraint violations. We formulate this safe reinforcement learning (RL) problem using the framework of a finite-horizon Constrained Markov Decision Process (CMDP) with an unknown transition probability function, where we model the safety requirements as constraints on the expected cumulative costs that must be satisfied during all episodes of learning. We propose a model-based safe RL algorithm that we call Doubly Optimistic and Pessimistic Exploration (DOPE), and show that it achieves an objective regret $$\tilde{O}(|\mathcal{S}|\sqrt{|\mathcal{A}| K})$$ without violating the safety constraints during learning, where $$|\mathcal{S}|$$ is the number of states, $$|\mathcal{A}|$$ is the number of actions, and $$K$$ is the number of learning episodes. Our key idea is to combine a reward bonus for exploration (optimism) with a conservative constraint (pessimism), in addition to the standard optimistic model-based exploration. DOPE is not only able to improve the objective regret bound, but also shows a significant empirical performance improvement as compared to earlier optimism-pessimism approaches.more » « less
-
null (Ed.)Inspired by the demands of real-time climate and weather forecasting, we develop optimistic online learning algorithms that require no parameter tuning and have optimal regret guarantees under delayed feedback. Our algorithms -- DORM, DORM+, and AdaHedgeD -- arise from a novel reduction of delayed online learning to optimistic online learning that reveals how optimistic hints can mitigate the regret penalty caused by delay. We pair this delay-as-optimism perspective with a new analysis of optimistic learning that exposes its robustness to hinting errors and a new meta-algorithm for learning effective hinting strategies in the presence of delay. We conclude by benchmarking our algorithms on four subseasonal climate forecasting tasks, demonstrating low regret relative to state-of-the-art forecasting models.more » « less
-
Meila, Marina; Zhang, Tong (Ed.)Inspired by the demands of real-time climate and weather forecasting, we develop optimistic online learning algorithms that require no parameter tuning and have optimal regret guarantees under delayed feedback. Our algorithms—DORM, DORM+, and AdaHedgeD—arise from a novel reduction of delayed online learning to optimistic online learning that reveals how optimistic hints can mitigate the regret penalty caused by delay. We pair this delay-as-optimism perspective with a new analysis of optimistic learning that exposes its robustness to hinting errors and a new meta-algorithm for learning effective hinting strategies in the presence of delay. We conclude by benchmarking our algorithms on four subseasonal climate forecasting tasks, demonstrating low regret relative to state-of-the-art forecasting models.more » « less
-
We present Q-functionals, an alternative architecture for continuous control deep reinforcement learning. Instead of returning a single value for a state-action pair, our network transforms a state into a function that can be rapidly evaluated in parallel for many actions, allowing us to efficiently choose high-value actions through sampling. This contrasts with the typical architecture of off-policy continuous control, where a policy network is trained for the sole purpose of selecting actions from the Q-function. We represent our action-dependent Q-function as a weighted sum of basis functions (Fourier, Polynomial, etc) over the action space, where the weights are state-dependent and output by the Q-functional network. Fast sampling makes practical a variety of techniques that require Monte-Carlo integration over Q-functions, and enables action-selection strategies besides simple value-maximization. We characterize our framework, describe various implementations of Q-functionals, and demonstrate strong performance on a suite of continuous control tasks.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- The DOI is not currently available.
