LGAIJun 4, 2025

Horizon Reduction Makes RL Scalable

arXiv:2506.04168v340 citationsh-index: 43Has Code
Originality Incremental advance
AI Analysis

This addresses a scalability problem for offline RL practitioners, offering incremental improvements through horizon reduction techniques.

The paper tackles the poor scalability of offline reinforcement learning (RL) algorithms on complex tasks with large datasets, finding that long horizons are a fundamental barrier and showing that horizon reduction techniques, including a new method called SHARSA, substantially improve scalability and achieve the best asymptotic performance.

In this work, we study the scalability of offline reinforcement learning (RL) algorithms. In principle, a truly scalable offline RL algorithm should be able to solve any given problem, regardless of its complexity, given sufficient data, compute, and model capacity. We investigate if and how current offline RL algorithms match up to this promise on diverse, challenging, previously unsolved tasks, using datasets up to 1000x larger than typical offline RL datasets. We observe that despite scaling up data, many existing offline RL algorithms exhibit poor scaling behavior, saturating well below the maximum performance. We hypothesize that the horizon is the main cause behind the poor scaling of offline RL. We empirically verify this hypothesis through several analysis experiments, showing that long horizons indeed present a fundamental barrier to scaling up offline RL. We then show that various horizon reduction techniques substantially enhance scalability on challenging tasks. Based on our insights, we also introduce a minimal yet scalable method named SHARSA that effectively reduces the horizon. SHARSA achieves the best asymptotic performance and scaling behavior among our evaluation methods, showing that explicitly reducing the horizon unlocks the scalability of offline RL. Code: https://github.com/seohongpark/horizon-reduction

Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes