On the Importance of Multistability for Horizon Generalization in Reinforcement Learning

• URL: http://arxiv.org/abs/2605.12206v1
• Date: Tue, 12 May 2026 14:45:55 GMT
• Title: On the Importance of Multistability for Horizon Generalization in Reinforcement Learning
• Abstract summary: Long-horizon Markov decision processes (POMDPs) are particularly challenging.<n>Training suffers from poor generalization, severe sample inefficiency, and prohibitive exploration costs.<n>We formalized temporal horizon generalization and experimentally evaluated the ability of nonlinear and parallelizable RNN variants to achieve it.
• Score: 0.0
• License: http://creativecommons.org/licenses/by-nc-sa/4.0/
• Abstract: In reinforcement learning (RL), agents acting in partially observable Markov decision processes (POMDPs) must rely on memory, typically encoded in a recurrent neural network (RNN), to integrate information from past observations. Long-horizon POMDPs, in which the relevant observation and the optimal action are separated by many time steps (called the horizon), are particularly challenging: training suffers from poor generalization, severe sample inefficiency, and prohibitive exploration costs. Ideally, an agent trained on short horizons would retain optimal behavior at arbitrarily longer ones, but no formal framework currently characterizes when this is achievable. To fill this gap, we formalized temporal horizon generalization, the property that a policy remains optimal for all horizons, derived a necessary and sufficient condition for it, and experimentally evaluated the ability of nonlinear and parallelizable RNN variants to achieve it. This paper presents the resulting theoretical framework, the empirical evaluation, and the dynamical interpretation linking RNN behavior to temporal horizon generalization. Our analyses reveal that multistability is necessary for temporal horizon generalization and, in simple tasks, sufficient; more complex tasks further require transient dynamics. In contrast, modern parallelizable architectures, namely state space models and gated linear RNNs, are monostable by construction and consequently fail to generalize across temporal horizons. We conclude that multistability and transient dynamics are two essential and complementary dynamical regimes for horizon generalization, and that no current parallelizable RNN exhibits both. Designing parallelizable architectures that combine these regimes thus emerges as a key direction for scalable long-horizon RL.

Related papers

• Forget Less, Generalize More: Unifying Temporal and Structural Adaptation for Dynamic Graphs [9.230468469629345]
  We propose a unified framework that maintains a retentive representation state encoding both temporal memory and structural context.<n>We show that Dual-Scale Retentive Dynamics consistently achieves state-of-the-art performance on both link prediction and node classification tasks.
  arXiv  Detail & Related papers  (2026-05-28T06:47:08Z)
• On the Generalization Behavior of Deep Residual Networks From a Dynamical System Perspective [1.0388986221727612]
  Deep neural networks (DNNs) have significantly advanced machine learning, with model depth playing a central role in their successes.<n>In this work, we establish generalization error bounds for both discrete- and continuous-time residual networks (ResNets) by combining Rademacher complexity, flow maps of dynamical systems, and the convergence behavior of ResNets in the deep-layer limit.<n>Findings provide a unified understanding of generalization across both discrete- and continuous-time ResNets, helping to close the gap in both the order of sample complexity and assumptions between the discrete- and continuous-time settings.
  arXiv  Detail & Related papers  (2026-02-24T13:59:06Z)
• OSDEnhancer: Taming Real-World Space-Time Video Super-Resolution with One-Step Diffusion [64.10689934231165]
  Video super-resolution models (DMs) have demonstrated exceptional success in video super-resolution (VSR)<n>Their potential for space-time video super-resolution (STVSR) necessitates recovering realistic visual content from low to high-resolution but also improving the frame rate with coherent dynamics.<n>We propose OSDEnhancer, a framework that represents the first method to initialize real-world STVSR through an efficient one-step diffusion process.<n> Experiments demonstrate that the proposed method achieves state-of-the-art performance while maintaining superior capability in real-world scenarios.
  arXiv  Detail & Related papers  (2026-01-28T06:59:55Z)
• Sample-Efficient Neurosymbolic Deep Reinforcement Learning [49.60927398960061]
  We propose a neuro-symbolic Deep RL approach that integrates background symbolic knowledge to improve sample efficiency.<n>Online reasoning is performed to guide the training process through two mechanisms.<n>We show improved performance over a state-of-the-art reward machine baseline.
  arXiv  Detail & Related papers  (2026-01-06T09:28:53Z)
• Nemotron-Cascade: Scaling Cascaded Reinforcement Learning for General-Purpose Reasoning Models [71.9060068259379]
  We propose cascaded domain-wise reinforcement learning to build general-purpose reasoning models.<n>Our 14B model, after RL, outperforms its SFT teacher, DeepSeek-R1-0528, on LiveCodeBench v5/v6 Pro and silver-medal performance in the 2025 International Olympiad in Informatics (IOI)
  arXiv  Detail & Related papers  (2025-12-15T18:02:35Z)
• RainDiff: End-to-end Precipitation Nowcasting Via Token-wise Attention Diffusion [64.49056527678606]
  We propose a Token-wise Attention integrated into not only the U-Net diffusion model but also the radar-temporal encoder.<n>Unlike prior approaches, our method integrates attention into the architecture without incurring the high resource cost typical of pixel-space diffusion.<n>Our experiments and evaluations demonstrate that the proposed method significantly outperforms state-of-the-art approaches, robustness local fidelity, generalization, and superior in complex precipitation forecasting scenarios.
  arXiv  Detail & Related papers  (2025-10-16T17:59:13Z)
• Transformer with Koopman-Enhanced Graph Convolutional Network for Spatiotemporal Dynamics Forecasting [12.301897782320967]
  TK-GCN is a two-stage framework that integrates geometry-aware spatial encoding with long-range temporal modeling.<n>We show that TK-GCN consistently delivers superior predictive accuracy across a range of forecast horizons.
  arXiv  Detail & Related papers  (2025-07-05T01:26:03Z)
• Universal In-Context Approximation By Prompting Fully Recurrent Models [86.61942787684272]
  We show that RNNs, LSTMs, GRUs, Linear RNNs, and linear gated architectures can serve as universal in-context approximators. We introduce a programming language called LSRL that compiles to fully recurrent architectures.
  arXiv  Detail & Related papers  (2024-06-03T15:25:13Z)
• Tractable Dendritic RNNs for Reconstructing Nonlinear Dynamical Systems [7.045072177165241]
  We augment a piecewise-linear recurrent neural network (RNN) by a linear spline basis expansion. We show that this approach retains all the theoretically appealing properties of the simple PLRNN, yet boosts its capacity for approximating arbitrary nonlinear dynamical systems in comparatively low dimensions.
  arXiv  Detail & Related papers  (2022-07-06T09:43:03Z)
• Provable Hierarchy-Based Meta-Reinforcement Learning [50.17896588738377]
  We analyze HRL in the meta-RL setting, where learner learns latent hierarchical structure during meta-training for use in a downstream task. We provide "diversity conditions" which, together with a tractable optimism-based algorithm, guarantee sample-efficient recovery of this natural hierarchy. Our bounds incorporate common notions in HRL literature such as temporal and state/action abstractions, suggesting that our setting and analysis capture important features of HRL in practice.
  arXiv  Detail & Related papers  (2021-10-18T17:56:02Z)

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.