[2603.29501] Target-Aligned Reinforcement Learning

Computer Science > Machine Learning arXiv:2603.29501 (cs) [Submitted on 31 Mar 2026] Title:Target-Aligned Reinforcement Learning Authors:Leonard S. Pleiss, James Harrison, Maximilian Schiffer View a PDF of the paper titled Target-Aligned Reinforcement Learning, by Leonard S. Pleiss and 2 other authors View PDF HTML (experimental) Abstract:Many reinforcement learning algorithms rely on target networks - lagged copies of the online network - to stabilize training. While effective, this mechanism introduces a fundamental stability-recency tradeoff: slower target updates improve stability but reduce the recency of learning signals, hindering convergence speed. We propose Target-Aligned Reinforcement Learning (TARL), a framework that emphasizes transitions for which the target and online network estimates are highly aligned. By focusing updates on well-aligned targets, TARL mitigates the adverse effects of stale target estimates while retaining the stabilizing benefits of target networks. We provide a theoretical analysis demonstrating that target alignment correction accelerates convergence, and empirically demonstrate consistent improvements over standard reinforcement learning algorithms across various benchmark environments. Subjects: Machine Learning (cs.LG); Artificial Intelligence (cs.AI) Cite as: arXiv:2603.29501 [cs.LG]   (or arXiv:2603.29501v1 [cs.LG] for this version)   https://doi.org/10.48550/arXiv.2603.29501 Focus to learn more arXiv-issued DOI via DataCite (pendi...