The Role of Environment Access in Agnostic Reinforcement Learning

• URL: http://arxiv.org/abs/2504.05405v1
• Date: Mon, 07 Apr 2025 18:19:56 GMT
• Title: The Role of Environment Access in Agnostic Reinforcement Learning
• Authors: Akshay Krishnamurthy, Gene Li, Ayush Sekhari,
• Abstract summary: We study Reinforcement Learning (RL) in environments with large state spaces.<n>We consider the weakest possible form of function approximation, called agnostic policy learning.<n>We show that sample-efficient policy learning is not possible in the standard online RL setting.
• Score: 37.457194209439926
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: We study Reinforcement Learning (RL) in environments with large state spaces, where function approximation is required for sample-efficient learning. Departing from a long history of prior work, we consider the weakest possible form of function approximation, called agnostic policy learning, where the learner seeks to find the best policy in a given class $\Pi$, with no guarantee that $\Pi$ contains an optimal policy for the underlying task. Although it is known that sample-efficient agnostic policy learning is not possible in the standard online RL setting without further assumptions, we investigate the extent to which this can be overcome with stronger forms of access to the environment. Specifically, we show that: 1. Agnostic policy learning remains statistically intractable when given access to a local simulator, from which one can reset to any previously seen state. This result holds even when the policy class is realizable, and stands in contrast to a positive result of [MFR24] showing that value-based learning under realizability is tractable with local simulator access. 2. Agnostic policy learning remains statistically intractable when given online access to a reset distribution with good coverage properties over the state space (the so-called $\mu$-reset setting). We also study stronger forms of function approximation for policy learning, showing that PSDP [BKSN03] and CPI [KL02] provably fail in the absence of policy completeness. 3. On a positive note, agnostic policy learning is statistically tractable for Block MDPs with access to both of the above reset models. We establish this via a new algorithm that carefully constructs a policy emulator: a tabular MDP with a small state space that approximates the value functions of all policies $\pi \in \Pi$. These values are approximated without any explicit value function class.

Related papers

• Confident Natural Policy Gradient for Local Planning in $q_π$-realizable Constrained MDPs [44.69257217086967]
  The constrained Markov decision process (CMDP) framework emerges as an important reinforcement learning approach for imposing safety or other critical objectives.<n>In this paper, we address the learning problem given linear function approximation with $q_pi$-realizability.
  arXiv  Detail & Related papers  (2024-06-26T17:57:13Z)
• The Power of Resets in Online Reinforcement Learning [73.64852266145387]
  We explore the power of simulators through online reinforcement learning with local simulator access (or, local planning) We show that MDPs with low coverability can be learned in a sample-efficient fashion with only $Qstar$-realizability. We show that the notorious Exogenous Block MDP problem is tractable under local simulator access.
  arXiv  Detail & Related papers  (2024-04-23T18:09:53Z)
• When is Agnostic Reinforcement Learning Statistically Tractable? [76.1408672715773]
  A new complexity measure, called the emphspanning capacity, depends solely on the set $Pi$ and is independent of the MDP dynamics. We show there exists a policy class $Pi$ with a bounded spanning capacity that requires a superpolynomial number of samples to learn. This reveals a surprising separation for learnability between generative access and online access models.
  arXiv  Detail & Related papers  (2023-10-09T19:40:54Z)
• General Policy Evaluation and Improvement by Learning to Identify Few But Crucial States [12.059140532198064]
  Learning to evaluate and improve policies is a core problem of Reinforcement Learning. A recently explored competitive alternative is to learn a single value function for many policies. We show that our value function trained to evaluate NN policies is also invariant to changes of the policy architecture.
  arXiv  Detail & Related papers  (2022-07-04T16:34:53Z)
• Offline Reinforcement Learning with Implicit Q-Learning [85.62618088890787]
  Current offline reinforcement learning methods need to query the value of unseen actions during training to improve the policy. We propose an offline RL method that never needs to evaluate actions outside of the dataset. This method enables the learned policy to improve substantially over the best behavior in the data through generalization.
  arXiv  Detail & Related papers  (2021-10-12T17:05:05Z)
• Minimax-Optimal Off-Policy Evaluation with Linear Function Approximation [49.502277468627035]
  This paper studies the statistical theory of batch data reinforcement learning with function approximation. Consider the off-policy evaluation problem, which is to estimate the cumulative value of a new target policy from logged history.
  arXiv  Detail & Related papers  (2020-02-21T19:20:57Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.