Nichol and Dhariwal: learned-variance and improved-objective diffusion training

What changed

Learned reverse-process variances and used modified objectives/noise schedules and importance sampling to improve diffusion likelihood and reduce required sampling steps.

Scientific scope

Learned reverse-process variances and used modified objectives/noise schedules and importance sampling to improve diffusion likelihood and reduce required sampling steps. Original DDPM, variational bounds and DDIM retain credit. Learned variance/objective changes are distinct from classifier guidance and later solver algorithms. Quality, likelihood and speed tradeoffs are benchmark dependent.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Original DDPM, variational bounds and DDIM retain credit. Learned variance/objective changes are distinct from classifier guidance and later solver algorithms. Quality, likelihood and speed tradeoffs are benchmark dependent.

Contributors

• Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.

• Improved Denoising Diffusion Probabilistic Models
• Nichol and Dhariwal: learned-variance and improved-objective diffusion training — author/institutional project

Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.

Full scientific record · Browse contribution grades · Request a correction