AI Paper Atlas

Research map / Label-efficient and robust learning / Self-supervised and contrastive

The SSL Interplay: Augmentations, Inductive Bias, and Generalization

Vivien Cabannes, Bobak Kiani, Randall Balestriero, Yann LeCun, Alberto Bietti

ICML 2023 · Label-efficient and robust learning › Self-supervised and contrastive · approach: Data-efficient and robust self-supervision · 7 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Add regularization lambda (lambda K^{-1} bias, i.e. T_lambda) to the spectral pretraining loss (ablation)VICReg-style population loss with kernel/architecture inductive bias · Theory: downstream excess risk on sphere S^7 polynomial targetsdownstream empirical risk (qualitative)Mixed
Use multiple augmentations per sample (multi-crop) versus more fresh inputs when total m x n is fixedPretraining with VICReg-style loss · Theory: pretraining convergencepretraining convergence rateNo clear effect
Negligible regularization (no inductive bias) in empirical pretraining with few augmentations (ablation)Empirical estimate of T_lambda · Gaussian kernel illustration (capacity trouble)recovery of cluster structure (qualitative)Hurt
Use stronger augmentations (reduce positive eigenspace of T_lambda) to avoid collapseSSL pretraining guidance · Practical guideline (theory-based, not experimentally verified)collapse (qualitative)No clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy