Challenge: MoE-based LLMs are not explicitly supervised to select suitable experts.
Approach: They propose Exploration-Driven Reinforcement Learning (ERL) which explicitly optimizes the router by exploration of alternative routing paths.
Outcome: The proposed method improves summarization (SAMSum, XSUM, question answering, and language modeling), and raises routing quality, delivering 8.9 higher MRR than baselines over 100 perturbed routing paths.

Similar Papers

A Closer Look into Mixture-of-Experts in Large Language Models (2025.findings-naacl)

Copied to clipboard

Challenge: Mixture-of-experts (MoE) architectures are gaining increasing attention for their unique properties and remarkable performance.
Approach: They propose a mixture-of-experts architecture that allows for model scaling without sacrificing computational efficiency.
Outcome: The proposed model increases model size without sacrificing computational efficiency . the proposed model is modular and can be used by a broad spectrum of practitioners .
Improved Policy Optimization for Mixture-of-Experts Models: Importance Sampling and Rewarding from an Expert-Centric Perspective (2026.findings-acl)

Copied to clipboard

Challenge: Existing approaches to reinforcement learning (RL) suffer from training instability . existing approaches often ignore token-specific discrepancies in expert assignments .
Approach: They propose to introduce expert-level importance sampling to reduce complexity of RL . they propose to leverage expert-centric granularity to ensure a rigorous alignment between reward signals and policy updates.
Outcome: The proposed method outperforms strong baselines across reasoning tasks.
Probing Semantic Routing in Large Mixture-of-Expert Models (2025.findings-emnlp)

Copied to clipboard

Challenge: large mixture-of-expert models have become increasingly common in the open domain . prior work has explored functional differentiation through routing behavior .
Approach: They investigate whether expert routing in large mixture-of-expert models is influenced by the semantics of the inputs.
Outcome: The results show that expert routing is influenced by the semantics of the inputs.
Harder Task Needs More Experts: Dynamic Routing in MoE Models (2024.acl-long)

Copied to clipboard

Challenge: Unlike existing MoE approaches that rely on fixed TopK Routing, our dynamic expert selection framework dynamically allocates experts based on the confidence level in expert selection for each input.
Approach: They propose a dynamic expert selection framework that dynamically allocates experts based on the confidence level in expert selection for each input.
Outcome: The proposed method achieves an average improvement of 0.7% with less than 90% activated parameters and outperforms dense models in QA and machine translation tasks.
On the Benefits of Learning to Route in Mixture-of-Experts Models (2023.emnlp-main)

Copied to clipboard

Challenge: Existing Mixture-of-Expert (MoE) models allow us to scale up model sizes while keeping the amount of compute time fixed.
Approach: They propose to use a router to route inputs to experts in a layer to scale up model sizes while keeping the amount of compute time fixed.
Outcome: The proposed model scales up with the help of a router that routes input tokens to experts in a layer and shows that it is more efficient than a non-trainable router.
Enhancing Efficiency and Exploration in Reinforcement Learning for LLMs (2025.emnlp-main)

Copied to clipboard

Challenge: Existing approaches allocate an equal number of rollouts to all questions during the RL process, which is inefficient.
Approach: They propose a mechanism for dynamically allocating rollout budgets based on the difficulty of the problems, enabling more efficient RL training.
Outcome: The proposed model improves response precision while preserving exploratory ability to uncover potential correct pathways.
Parameter-Efficient Routed Fine-Tuning: Mixture-of-Experts Demands Mixture of Adaptation Modules (2026.findings-eacl)

Copied to clipboard

Challenge: Existing Parameter-Efficient Fine-Tuning (PEFT) strategies that focus on specialized experts are not effective for Mixture-of-Experts (MoE).
Approach: They propose to integrate a dynamic routing mechanism among specialized experts in Mixture-of-Experts (MoE) .
Outcome: Extensive experiments on commonsense and math reasoning tasks validate the performance and efficiency of the proposed routed approach.
Part-Of-Speech Sensitivity of Routers in Mixture of Experts Models (2025.coling-main)

Copied to clipboard

Challenge: a study examines the behavior of routers in Mixture of Experts (MoE) models . experts with similar linguistic traits are often routed to the same expert regardless of context .
Approach: They investigate how tokens are routed based on their linguistic features . they aim to explore whether experts specialize in processing tokens with similar linguistic traits .
Outcome: The proposed model-integrated routers are based on Mixture of Experts (MoE) models . the results show that expert specialization is high for POS categories .
Exploring Domain Robust Lightweight Reward Models based on Router Mechanism (2024.findings-acl)

Copied to clipboard

Challenge: Recent advances in large language models have relied on the large reward model for fine-tuning, but the use of a single reward model across domains may not always be optimal.
Approach: They propose to use router mechanisms to train small language models in a domain-specific manner . they use internal routers, external routers and router adapters to create a single reward model .
Outcome: The proposed approach reduces parameter size while minimizing parameter size.
Towards Stable and Effective Reinforcement Learning for Mixture-of-Experts (2026.acl-long)

Copied to clipboard

Challenge: Reinforcement learning with verifiable rewards (RLVR) training with Mixture-of-Experts policies remains fragile and prone to reward collapse.
Approach: They propose a router shift-based policy optimization method that computes a per-token router-shift ratio conditioned on the previously activated experts and applies stop-gradient and a lower-bound floor.
Outcome: The proposed method achieves better performance and greater stability than previous methods.

What is GenGO?

GenGO is an NLP powered publication search system. It currenctly indexes 30k+ papers from ACL Anthology, and implements multi-aspect summarization, semantic search, and more!

Information

About
Limitations