Three Papers Accepted in ICML 2026

Three papers on efficient reasoning, sparse attention, and diffusion LLMs are accepted by ICML'2026.

The accepted papers are listed below:

  • Attention Sink Forges Native MoE in Attention Layers: Sink-Aware Training to Address Head Collapse (led by Zizhuo Fu)
  • TEAM: Temporal–Spatial Consistency Guided Expert Activation for MoE Diffusion Language Model Acceleration (led by Linye Wei)
  • HyPER: Bridging Exploration and Exploitation for Scalable LLM Reasoning with Hypothesis Path Expansion and Reduction (led by Shengxuan Qiu, Haochen Huang)