Metraipolis › archive › dc_f321c340bc24
Simplex Diffusion Models — arXiv:2609.35553
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.35553
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-07T22:39:13.922Z
- SHA-256 of the archived text
7840a87e9a81820acba5bb3b9ca588227eb619994472b8feedbdd58baf978995- Stable URL of this page
https://metraipolis.com/documento/dc_f321c340bc24
How to cite
Simplex Diffusion Models — arXiv:2609.35553 — arXiv (en), https://arxiv.org/abs/2609.35553 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_f321c340bc24 (Metraipolis archive dc_f321c340bc24, sha256 7840a87e9a81820acba5bb3b9ca588227eb619994472b8feedbdd58baf978995)
BibTeX
@misc{dcf321c340bc24,
title = {Simplex Diffusion Models — arXiv:2609.35553},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.35553},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_f321c340bc24; sha256 7840a87e9a81820acba5bb3b9ca588227eb619994472b8feedbdd58baf978995},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.35553. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Diffusion models have revolutionized generative modeling for continuous data through the gradual refinement of a belief state. This iterative refinement has not yet carried over to discrete diffusion models, which discard uncertainty at intermediate steps through categorical sampling (information collapse). We propose Simplex Diffusion Models (SDMs), a framework that lifts the diffusion process to the probability simplex to represent beliefs over categories. SDMs admit probability paths with closed-form reverse transitions and can be trained with a simple cross-entropy loss. Contrary to earlier proposals such as Dirichlet Flow Matching which requires integrating an ordinary differential equation, we introduce a DDIM-like sampler with a tunable level of stochasticity. Because SDMs operate on samples on the simplex, they can carry uncertainty across denoising steps, which mitigates information collapse. On OpenWebText, SDMs are competitive with strong Discrete Diffusion baselines, achieving $17.0$ GenPPL at $5.46$ unigram entropy in 64 sampling steps, close to real validation data. Even without Self-Conditioning (SC), SDMs outperform masked and uniform diffusion (with SC or predictor-corrector sampling) on code generation (TinyGSM, $T=0.1$; $49.0\%$ vs. $45.8\%$). Distilled down to 8 steps, SDMs solve $32.1\%$ of GSM8K problems, more than distilled Discrete Diffusion models with 128 steps ($21.4\%$). Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-29 in cs.LG). Categorie arXiv: Machine Learning (cs.LG) ; Machine Learning (stat.ML). Pagina del lavoro: https://arxiv.org/abs/2609.35553 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito). Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina. Letto il 2026-10-07T22:14:28.473Z dalla pagina https://arxiv.org/catchup/cs.LG/2026-09-29?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).