This paper explores the use of Maximum Causal Entropy Inverse Reinforcement Learning (IRL) within the context of discrete-time stationary Mean-Field Games (MFGs) characterized by finite state spaces and an infinite-horizon, discounted-reward setting. Although the resulting optimization problem is non-convex with respect to policies, we reformulate it as a convex optimization problem in terms of state-action occupation measures by leveraging the linear programming framework of Markov Decision Processes. Based on this convex reformulation, we introduce a gradient descent algorithm with a guaranteed convergence rate to efficiently compute the optimal solution. Moreover, we develop a new method that conceptualizes the MFG problem as a Generalized Nash Equilibrium Problem (GNEP), enabling effective computation of the mean-field equilibrium for forward reinforcement learning (RL) problems and marking an advancement in MFG solution techniques. We further illustrate the practical applicability of our GNEP approach by employing this algorithm to generate data for numerical MFG examples.

We consider learning approximate Nash equilibria for discrete-time mean-field games with stochastic nonlinear state dynamics subject to both average and discounted costs. To this end, we introduce a mean-field equilibrium (MFE) operator, whose fixed point is a mean-field equilibrium, i.e., equilibrium in the infinite population limit. We first prove that this operator is a contraction, and propose a learning algorithm to compute an approximate mean-field equilibrium by approximating the MFE operator with a random one. Moreover, using the contraction property of the MFE operator, we establish the error analysis of the proposed learning algorithm. We then show that the learned mean-field equilibrium constitutes an approximate Nash equilibrium for finite-agent games.

Reinforcement learning algorithms often require finiteness of state and action spaces in Markov decision processes (MDPs) (also called controlled Markov chains) and various efforts have been made in the literature towards the applicability of such algorithms for continuous state and action spaces. In this paper, we show that under very mild regularity conditions (in particular, involving only weak continuity of the transition kernel of an MDP), Q-learning for standard Borel MDPs via quantization of states and actions (called Quantized Q-Learning) converges to a limit, and furthermore this limit satisfies an optimality equation which leads to near optimality with either explicit performance bounds or which are guaranteed to be asymptotically optimal. Our approach builds on (i) viewing quantization as a measurement kernel and thus a quantized MDP as a partially observed Markov decision process (POMDP), (ii) utilizing near optimality and convergence results of Q-learning for POMDPs, and (iii) finally, near-optimality of finite state model approximations for MDPs with weakly continuous kernels which we show to correspond to the fixed point of the constructed POMDP. Thus, our paper presents a very general convergence and approximation result for the applicability of Q-learning for continuous MDPs.

This paper studies fixed-rate randomized vector quantization under the constraint that the quantizer's output has a given fixed probability distribution. A general representation of randomized quantizers that includes the common models in the literature is introduced via appropriate mixtures of joint probability measures on the product of the source and reproduction alphabets. Using this representation and results from optimal transport theory, the existence of an optimal (minimum distortion) randomized quantizer having a given output distribution is shown under various conditions. For sources with densities and the mean square distortion measure, it is shown that this optimum can be attained by randomizing quantizers having convex codecells. For stationary and memoryless source and output distributions, a rate-distortion theorem is proved, providing a single-letter expression for the optimum distortion in the limit of large blocklengths.

This paper studies a Shannon-theoretic version of the generalized distribution preserving quantization problem where a stationary and memoryless source is encoded subject to a distortion constraint and the additional requirement that the reproduction also be stationary and memoryless with a given distribution. The encoder and decoder are stochastic and assumed to have access to independent common randomness. Recent work has characterized the minimum achievable coding rate at a given distortion level when unlimited common randomness is available. Here, we consider the general case where the available common randomness may be rate limited. Our main result completely characterizes the set of achievable coding and common randomness rate pairs at any distortion level, thereby providing the optimal tradeoff between these two rate quantities. We also consider two variations of this problem where we investigate the effect of relaxing the strict output distribution constraint and the role of private randomness used by the decoder on the rate region. Our results have strong connections with Cuff's recent work on distributed channel synthesis. In particular, our achievability proof combines a coupling argument with the approach developed by Cuff, where instead of explicitly constructing the encoder-decoder pair, a joint distribution is constructed from which a desired encoder-decoder pair is established. We show, however, that for our problem, the separated solution of first finding an optimal channel and then synthesizing this channel results in a suboptimal rate region.

We consider the problem of optimal randomized vector quantization under a constraint on the output's distribution. The problem is formalized by introducing a general representation of randomized quantization via probability measures over the space of joint distributions on the source and reproduction alphabets. Using this representation and results from optimal transport theory, we show the existence of an optimal (minimum distortion) randomized quantizer having a fixed output distribution under various conditions. For sources with densities and the mean square distortion measure, we show that this optimum can be attained by randomizing quantizers having convex code cells. We also consider a relaxed version of the problem where the output marginal must belong to some neighborhood (in the weak topology) of a fixed probability measure. We demonstrate that finitely randomized quantizers form an optimal class for the relaxed problem.