ICML 2026: Self-Aware Scheduling (SAS) for Diffusion LMs
furongh · x · 2026-07-06
In a paper published at ICML 2026, researchers proposed the Self-Aware Scheduling (SAS) method to optimize reasoning in diffusion language models. The core idea: rather than forcing the model to "think longer," it is better to let the model know "which token to determine first."
By introducing a self-aware scheduling mechanism into diffusion LMs, SAS enables the model to dynamically decide the token generation order, thereby improving reasoning quality. This approach diverges from the mainstream CoT/long chain-of-thought route, offering a novel diffusion model perspective for LLM reasoning.
Related event: ICML 2026 Paper SAS: Optimizing Thought Scheduling in Diffusion LMs(15 posts)→
More from Research
- HarnessMM cuts agent tokens per task by up to 88% while raising success rates — omarsar0 · 2026-10-26
- Ben Recht: The world needs a truly open language model, and it's easier to build than you think — beenwrekt · 2026-10-26
- New Blog Breaks Black-Box Distillation Into Five Practical Steps — chrisrohlf · 2026-10-26
- DetectorScope turns black-box AI text detectors into auditable ones with sparse autoencoders — _akhaliq · 2026-10-26
- OPST paper shows LLMs self-improve with zero human data, guided only by topic descriptions — _akhaliq · 2026-10-26
- Tencent's HiSTEVE 1.0: hierarchical VLA agent plays Minecraft with native keyboard and mouse, hits 85% SOTA — _akhaliq · 2026-10-26