arXiv:2511. 22581v5 Announce Type: replace Abstract: We prove that in any Dec-POMDP, sufficiently high entropy regularization ensures that the policy gradient flow with tabular softmax parametrization always converges, for any initialization, to the same joint policy, and that this joint policy is equivariant w.
arXiv:2608.24488v1 Announce Type: new Abstract: Many continuous-control policies are optimized as unbounded Gaussians and then mapped into bounded actions. We show that where entropy is measured chan...
The paper investigates the problem of sharing a single critic across multiple parallel environments in reinforcement learning. It shows that when environments assign different expected returns to the same state, a shared critic must reconcile conflicting value targets, which can distort advantage estimates and misguide policy updates. The authors propose a simple fix—providing the critic with the environment index—demonstrating through bandit models and experiments on CartPole, MuJoCo, BipedalWalker, and 16 Procgen games that this conditional critic stabilizes learning and boosts returns, achieving a 40.8% improvement in aggregate normalized return on unseen levels.
arXiv:2608. 01425v1 Announce Type: cross Abstract: Training LLM-based multi-agent systems with multi-agent reinforcement learning is rapidly gaining traction, and a parallel line of work argues that such systems should be judged by their behavior, not only their reward.
arXiv:2607. 22186v2 Announce Type: replace Abstract: Asynchronous reinforcement learning (RL) accelerates large language model (LLM) post-training by overlapping rollout generation with policy optimization, but the resulting stale, off-policy data can destabilize optimization and ultimately cause policy collapse.
arXiv:2607. 08925v1 Announce Type: new Abstract: Training reinforcement-learning agents directly on physical robots makes every fall costly, since a fall can damage the platform and cannot be undone like a simulator reset; the goal is therefore to minimize falls during training rather than trade them off against return, as constrained Markov decision process (MDP) formulations do.
arXiv:2606. 01561v1 Announce Type: new Abstract: Aligning Large Language Models (LLMs) with human preferences is often formulated via Direct Preference Optimization (DPO).
arXiv:2609.36945v1 Announce Type: new Abstract: We study the learning dynamics of fine-tuning a policy model on self-generated and reward-weighted data, with particular focus on a generalized version...
arXiv:2608. 04149v1 Announce Type: cross Abstract: Swap regret governs the rate at which uncoupled learning dynamics converge to correlated equilibria in multiplayer general-sum games.
arXiv:2610.02198v1 Announce Type: cross Abstract: Several state-of-the-art methods for online reinforcement learning in continuous control improve policies using action gradients of a learned critic....
The paper investigates whether causal softmax attention can realize policy mirror descent (PMD) as a repeated controller rather than a one‑step algebraic identity. It constructs a fixed causal‑softmax actor–environment–one‑step‑critic protocol, detailing actor, routing, sampling, and normalization residuals, and shows that a frozen one‑step audit model closely approximates PMD. Empirical results demonstrate that the learned actor with an exact one‑step critic achieves median policy loss only about 5% higher than the exact PMD oracle across multiple control settings.
A deep network's loss is invariant to continuous symmetries of its parameters: the logit shift, the ReLU rescaling, the LayerNorm scale, the per-head attention rotation. Adam's per-coordinate preconditioner drifts along each symmetry orbit, which pulls the trajectory off the symmetry quotient where the optimization lives and blurs the singular-learning rate the quotient makes readable.
PokaiTrainer is a competitive Pokémon VGC agent that scales belief‑state search to handle simultaneous, large joint action spaces and stochastic outcomes. The system uses PokaiEngine, a Rust battle engine that efficiently enumerates joint action outcomes with high accuracy, and adapts Student of Games to solve each decision as a Bayesian matrix game under a compute budget. In live Showdown play, the agent achieved a 59% win rate against a human field averaging ~1320 Elo, reaching an Elo band of 1350‑1400 and briefly entering the top 500 of the format.