Simplex Diffusion Models proposed to fix information collapse in discrete diffusion
msalbergo · x · 2026-09-30
A thread introduces Simplex Diffusion Models, arguing that standard discrete diffusion models suffer from "information collapse" because they discard uncertainty at intermediate steps by sampling categorical tokens. The proposed simplex-based approach aims to address this; full technical details are in the original thread.
Related event: Simplex Diffusion Models Tackle Information Collapse in Discrete Diffusion(2 posts)→
More from Research
- VisionHOPE: First Visual Backbone Formulated as a Self-Modifying Learning System — CASIA · 2026-09-30
- Microsoft Trains a 'Night Science' Agent with RL, Expanding Research Directions 27.8% — MicrosoftResearch · 2026-09-30
- SentZero: sentence-centric VL pretraining boosts zero-shot multi-task chest X-ray analysis — Hangyul Yoon · 2026-09-30
- 500K Names Tested: Unequal Tokenizer Support Skews LLM Judgments in Hiring and Lending — Mir Tafseer Nayeem · 2026-09-30
- Training Kimi on 1,465 PDF tasks more than doubles general task success — echen · 2026-09-30
- 5,664 single-word teacher answers lift student +5.34pp on HumanEval+ without any code — _akhaliq · 2026-09-30