English

Hierarchical Mixture-of-Experts with Two-Stage Optimization

Machine Learning 2026-05-12 v1 Artificial Intelligence Optimization and Control

Abstract

Sparse Mixture-of-Experts (MoE) models scale capacity by routing each token to a small subset of experts. However, their routers exhibit a fundamental trade-off: strong load balancing can suppress expert specialization, while aggressive diversity often causes routing collapse. We propose Hi-MoE, a grouped MoE framework that decomposes routing control into two coupled levels: (i) inter-group balancing that enforces fair traffic across expert groups, and (ii) intra-group specialization that promotes complementary expert behaviors while preventing within-group collapse. Our analysis provides a principled explanation of how our hierarchical objectives reshape the router, thereby promoting stable specialization and mitigating collapse. We observe consistent improvements over recent sparse-routing and grouped-MoE baselines across NLP and vision benchmarks, and confirm robustness via scaling studies (model size, expert count) and targeted ablations. In large-scale pre-training on 58B tokens, Hi-MoE-7B achieves a 5.6% perplexity reduction and a 40% improvement in expert balance over OLMoE-7B across diverse evaluation domains.

Keywords

Cite

@article{arxiv.2605.08292,
  title  = {Hierarchical Mixture-of-Experts with Two-Stage Optimization},
  author = {Gleb Molodtsov and Alexander Miasnikov and Aleksandr Beznosikov},
  journal= {arXiv preprint arXiv:2605.08292},
  year   = {2026}
}
R2 v1 2026-07-01T12:58:41.632Z