NeurIPS 2026 paper shows noise-conditioning-free diffusion models implicitly implement Riemannian gradient flow
docmilanfar · x · 2026-09-25
The author's sole submission to NeurIPS 2026 has been accepted. The paper tackles a counterintuitive finding: diffusion models typically require exact noise level schedules, yet some recent models work without explicit noise conditioning. The authors show that these time-invariant fields implicitly implement a Riemannian gradient flow on an energy landscape, explaining how denoising succeeds without explicit time/noise conditioning.
More from Research
- GRAM accepted to NeurIPS 2026: recursive reasoning becomes stochastic latent trajectories — SungjinAhn_ · 2026-09-25
- Photonic matrix core on thin-film lithium niobate runs in-situ backpropagation at 8-bit precision — jwt0625 · 2026-09-25
- Agentic robotics keeps bolting GPT-6 onto harnesses — but where are systems that learn from physical failure? — DJiafei · 2026-09-25
- Dev Proposes "Evidentiality Framework": Provenance Markers for Every LLM Claim — jzesbaugh · 2026-09-25
- New Paper Quantifies When Rubric-Based RL Genuinely Improves Models vs. Hacks the Verifier — heghbalz · 2026-09-25
- DeepSeek's DSec sandbox platform serves 3M sandboxes daily for agent RL training; Liang Wenfeng co-authors — teortaxesTex · 2026-09-25