Chen and He: SimSiam stop-gradient visual self-supervision

What changed

Showed that a predictor and stop-gradient operation can train a simple shared-weight Siamese representation learner without negative pairs, momentum encoders or unusually large batches.

Scientific scope

Showed that a predictor and stop-gradient operation can train a simple shared-weight Siamese representation learner without negative pairs, momentum encoders or unusually large batches. Siamese and positive-pair learning have predecessors. Stop-gradient's empirical role and an alternating-optimization hypothesis are not a general proof against representation collapse. Original2020 preprint precedes CVPR2021.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Siamese and positive-pair learning have predecessors. Stop-gradient's empirical role and an alternating-optimization hypothesis are not a general proof against representation collapse. Original2020 preprint precedes CVPR2021.

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.

• Exploring Simple Siamese Representation Learning
• Chen and He: SimSiam stop-gradient visual self-supervision — author/institution implementation

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