Expert-Space Exploration in MoE Reinforcement Learning
ESRL explores MoE expert-routing space during RL, lifting Qwen3-30B-A3B Pass@1 by 3.2 points over GRPO at no extra cost.
ESRL (Expert-Space Exploration Reinforcement Learning) is an architecture-aware framework that treats expert routing in Mixture-of-Experts models as an additional source of rollout diversity. It preserves high-confidence experts as anchors, restricts stochastic routing to a plausible candidate pool, adapts perturbation strength to router entropy, and replays recorded expert paths during policy optimization. ESRL achieves the best performance across top-K, top-1, and shared-expert routing backbones on mathematics, science, and code tasks; on Qwen3-30B-A3B it improves average Pass@1 and Pass@8 over GRPO by 3.2 and 4.5 percentage points respectively, without additional sampling or compute.
- Routing perturbation adds rollout diversity like higher decoding temperature.
- Anchoring high-confidence experts avoids quality degradation from bad routing.
- Qwen3-30B-A3B: +3.2 Pass@1 and +4.5 Pass@8 over GRPO.
- Works across top-K, top-1, and shared-expert MoE routing at no extra cost.
Full article250 words · extracted from arxiv.org · click to collapse
Reinforcement learning (RL) has become central to post-training of large language models. Recent advances in RL for Mixture-of-Experts (MoE) models have primarily focused on improving optimization stability and training efficiency, while treating the expert selection as a fixed component. Since routing determines the sparse computation paths that induce output distributions, expert selection offers an additional source of rollout diversity. Through empirical analysis, we find that perturbing expert routing effectively alters model output and increases rollout diversity, which is similar to increasing the decoding temperature. However, direct perturbation can activate unsuitable experts and substantially degrade rollout quality. Motivated by these observations, we introduce Expert-Space Exploration Reinforcement Learning (ESRL), an architecture-aware framework that explicitly explores the expert-routing space of MoE models. ESRL preserves high-confidence experts as anchors, and restricts stochastic routing to a plausible candidate pool, thereby retaining reliable computation paths. The perturbation strength is further adapted according to router entropy to avoid over-perturbation. To mitigate the routing mismatch introduced by perturbation, ESRL records the expert paths used during rollout and replays them during policy optimization. Experiments demonstrate that ESRL achieves the best performance across MoE backbones with top-K, top-1, and shared-expert routing, as well as across mathematics, science, and code tasks without additional sampling or computational cost. Specifically, ESRL on Qwen3-30B-A3B achieves the best among all compared methods, improving average Pass@1 and Pass@8 over GRPO by 3.2 and 4.5 percentage points, respectively. Further analyses of expert utilization and training dynamics provide insights into how exploiting MoE-specific routing structure benefits RL training.
Text extracted automatically; images, tables and formatting may be missing. Original: https://arxiv.org/abs/2609.13058