Policy Gradient

We study infinite-horizon Constrained Markov Decision Processes (CMDPs) with general policy parameterizations and multi-layer neural network critics. Existing theoretical analyses for constrained reinforcement learning largely rely on tabular policies or linear critics, which limits their applicability to high-dimensional and continuous control problems. We propose a primal-dual natural actor-critic algorithm that integrates neural critic estimation with natural policy gradient updates and leverages Neural Tangent Kernel (NTK) theory to control function-approximation error under Markovian sampling, without requiring access to mixing-time oracles. We establish global convergence and cumulative constraint violation rates of $\tilde{\mathcal{O}}(T^-1/4)$ up to approximation errors induced by the policy and critic classes. Our results provide the first such guarantees for CMDPs with general policies and multi-layer neural critics, substantially extending the theoretical foundations of actor-critic methods beyond the linear-critic regime.

We study post-training linear autoregressive models with outcome and process rewards. Given a context $\boldsymbol{x}$, the model must predict the response $\boldsymbol{y} \in Y^N$, a sequence of length $N$ that satisfies a $γ$ margin condition, an extension of the standard separability to sequences. We prove that on test samples where the base model achieves a non-trivial likelihood $α$, a variant of policy gradient (PG) can achieve likelihood $1 - \varepsilon$ with an essentially minimax optimal number of reward queries $\tilde{O}((α^{-1} + \varepsilon^{-1})/γ^2)$. However, a barrier arises for going beyond the support of the base model. We prove that the overall expected error after post-training with outcome rewards is governed by a property of the base model called the Likelihood Quantile (LQ), and that variants of PG, while minimax optimal, may require a number of reward queries exponential in $N$ to go beyond this support, regardless of the pre-training algorithm. To overcome this barrier, we study post-training with a process reward model, and demonstrate how PG variants in this setting avoid the curse of dimensionality in $N$ via dependence on a token-level LQ. Along the way, we prove that under the margin condition, SGD with adaptive learning rate (LR) achieves a near optimal test error for statistical learning, and PG with adaptive LR achieves a near optimal number of mistakes for online learning while being computationally efficient whenever possible, both of which may be of independent interest.

We investigate the theoretical aspects of offline reinforcement learning (RL) under general function approximation. While prior works (e.g., Xie et al., 2021) have established the theoretical foundations of learning a good policy from offline data via pessimism, existing algorithms that are computationally tractable (often in an oracle-efficient sense), such as PSPI, only apply to finite and small action spaces. Moreover, these algorithms rely on state-wise mirror descent and require actors to be implicitly induced from the critic functions, failing to accommodate standalone policy parameterization which is ubiquitous in practice. In this work, we address these limitations and extend the theoretical guarantees to parameterized policy classes over large or continuous action spaces. When extending mirror descent to parameterized policies, we identify contextual coupling as the core difficulty, and show how connecting mirror descent to natural policy gradient leads to novel analyses, guarantees, and algorithmic insights, including a surprising unification between offline RL and imitation learning.

This paper studies networked multi-agent reinforcement learning (NMARL) with interdependent rewards and coupled policies. In this setting, each agent's reward depends on its own state-action pair as well as those of its direct neighbors, and each agent's policy is parameterized by its local parameters together with those of its $κ_{p}$-hop neighbors, with $κ_{p}\geq 1$ denoting the coupled radius. The objective of the agents is to collaboratively optimize their policies to maximize the discounted average cumulative reward. To address the challenge of interdependent policies in collaborative optimization, we introduce a novel concept termed the ``neighbors' averaged $Q$-function'' and derive a new expression for the coupled-policy gradient. Based on these theoretical foundations, we develop a distributed scalable coupled policy (DSCP) algorithm, where each agent relies only on the state-action pairs of its $κ_{p}$-hop neighbors and the rewards of its $(κ_{p}+1)$-hop neighbors. Specifically, in the DSCP algorithm, we employ a geometric 2-horizon sampling method that does not require storing a full $Q$-table to obtain an unbiased estimate of the coupled policy gradient. Moreover, each agent interacts exclusively with its direct neighbors to obtain accurate policy parameters, while maintaining local estimates of other agents' parameters to execute its local policy and collect samples for optimization. These estimates and policy parameters are updated via a push-sum protocol, enabling distributed coordination of policy updates across the network. We prove that the joint policy produced by the proposed algorithm converges to a first-order stationary point of the objective function. Finally, the effectiveness of DSCP algorithm is demonstrated through simulations in a robot path planning environment, showing clear improvement over state-of-the-art methods.

Multi-agent games in dynamic nonlinear settings are challenging due to the time-varying interactions among the agents and the non-stationarity of the (potential) Nash equilibria. In this paper we consider model-free games, where agent transitions and costs are observed without knowledge of the transition and cost functions that generate them. We propose a novel distributed policy structure that follows the communication constraints in multi-team games, with multiple agents per team, and learned through policy gradients. Our formulation is inspired by the structure of distributed policies in linear quadratic games, which take the form of time-varying linear feedback gains. In the nonlinear case, we model the policies as nonlinear feedback gains, parameterized by self-attention layers to account for the time-varying multi-agent communication topology. We demonstrate that our approach achieves strong performance in several settings, including distributed linear and nonlinear regulation, and simulated and real multi-robot pursuit-and-evasion games.

Policy gradient (PG) methods have played an essential role in the empirical successes of reinforcement learning. In order to handle large state-action spaces, PG methods are typically used with function approximation. In this setting, the approximation error in modeling problem-dependent quantities is a key notion for characterizing the global convergence of PG methods. We study Softmax PG with linear function approximation (referred to as $\texttt{Lin-SPG}$) and demonstrate that the approximation error is irrelevant to the algorithm's global convergence even in the bandit setting. Consequently, we rethink the effect of approximation error in the standard stochastic multi-armed bandit problem. We first identify the conditions on the policy feature representation that can guarantee the asymptotic global convergence of $\texttt{Lin-SPG}$. Under these feature conditions, we further prove that $T$ iterations of $\texttt{Lin-SPG}$ with a problem-specific learning rate result in an $O(1/T)$ convergence to the optimal policy. Moreover, we prove that $\texttt{Lin-SPG}$ with an arbitrary constant learning rate can ensure asymptotic convergence to the optimal policy.

Policy gradient methods are among the most successful approaches for solving challenging reinforcement learning problems. Despite their empirical success, many state-of-the-art policy gradient algorithms for discounted problems estimate update directions using a sampling distribution that differs from the discounted state distribution required by the policy gradient theorem. In this work, we analyze the impact of this distribution mismatch on policy gradient methods. Specifically, {for tabular direct and softmax parameterizations, we establish global convergence of objective values under biased updates}. {For tabular softmax policies, we further derive the exact leading-order coefficients of the asymptotic objective errors under biased and normalized exact updates. Their ratio provides an explicit criterion for determining when distribution mismatch accelerates or slows asymptotic convergence.} We then extend the analysis to more general policy parameterizations by deriving explicit bounds on both the state distribution mismatch and the induced normalized policy-gradient mismatch in episodic and continuing MDPs. These mismatches vanish at least linearly as the discount factor approaches one. Under an additional smoothness assumption, we further show that the limit inferior of the squared norm of the normalized exact policy-gradient direction along the biased iterates is bounded by $O((1-γ)^2)$ in both settings. Our results provide theoretical insights into the robustness of policy gradient methods to distribution mismatch and shed light on the gap between their theoretical formulations and practical implementations.

Kakade's natural policy gradient method has been studied extensively in recent years, showing linear convergence with and without regularization. We study another natural gradient method based on the Fisher information matrix of the state-action distributions which has received little attention from the theoretical side. Here, the state-action distributions follow the Fisher-Rao gradient flow inside the state-action polytope with respect to a linear potential. Therefore, we study Fisher-Rao gradient flows of linear programs more generally and show linear convergence with a rate that depends on the geometry of the linear program. Equivalently, this yields an estimate on the error induced by entropic regularization of the linear program which improves existing results. We extend these results and show sublinear convergence for perturbed Fisher-Rao gradient flows and natural gradient flows up to an approximation error. In particular, these general results cover the case of state-action natural policy gradients.

The policy gradient theorem describes the gradient of the expected discounted return with respect to an agent's policy parameters. However, many policy gradient methods omit the outer discount factor, yielding an update that need not equal the gradient of the discounted objective. Is the resulting update the gradient of another objective? This has been an open question for several years, and this lack of theoretical clarity has led to an abundance of misstatements in the literature. We answer this question for the resulting episode-summed, on-policy update by proving that it need not equal the gradient of any function. Further, we construct a counterexample in which exact expected updates drive the policy toward a deterministic limit that is globally pessimal with respect to both the discounted and undiscounted objectives, while the parameter diverges. We motivate this work by surveying the literature and showing that there remains a widespread misunderstanding regarding discounted policy gradient methods, with errors present even in highly-cited papers published at top conferences.