EDM2: magnitude-controlled diffusion training and post-hoc averaging

What changed

Redesigned denoiser operations and weight updates to reduce expected-magnitude drift and imbalances, and reconstructed alternative averaging profiles from saved averaged parameter snapshots after training.

Scientific scope

Redesigned denoiser operations and weight updates to reduce expected-magnitude drift and imbalances, and reconstructed alternative averaging profiles from saved averaged parameter snapshots after training. Magnitude arguments use statistical assumptions, not exact preservation for arbitrary correlated activations. Post-hoc averaging is a least-squares approximation using stored snapshots, not exact recovery of any history without storage. Inherits diffusion/ADM and normalization ideas; original report is2023, not the later conference year.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Magnitude arguments use statistical assumptions, not exact preservation for arbitrary correlated activations. Post-hoc averaging is a least-squares approximation using stored snapshots, not exact recovery of any history without storage. Inherits diffusion/ADM and normalization ideas; original report is2023, not the later conference year.

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.

• Analyzing and Improving the Training Dynamics of Diffusion Models

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