paper-with-me

홈 › Papers

Feature Normalization Prevents Collapse of Non-contrastive Learning Dynamics

2023-09-28 · Han Bao

Contrastive learning is a self-supervised representation learning framework, where two positive views generated through data augmentation are made similar by an attraction force in a data representation space, while a repulsive force makes them far from negative examples. Non-contrastive learning, represented by BYOL and SimSiam, further gets rid of negative examples and improves computational efficiency. While learned representations may collapse into a single point due to the lack of the repulsive force at first sight, Tian et al. (2021) revealed through the learning dynamics analysis that the representations can avoid collapse if data augmentation is sufficiently stronger than regularization. However, their analysis does not take into account commonly-used feature normalization, a normalizer before measuring the similarity of representations, and hence excessively strong regularization may collapse the dynamics, which is an unnatural behavior under the presence of feature normalization. Therefore, we extend the previous theory based on the L2 loss by considering the cosine loss, which involves feature normalization. We show that the cosine loss induces sixth-order dynamics (while the L2 loss induces a third-order one), in which a stable equilibrium dynamically emerges even if there are only collapsed solutions with given initial parameters. Thus, we offer a new understanding that feature normalization plays an important role in robustly preventing the dynamics collapse.

📄 PDF Abstract BibTeX arXiv:2309.16109

Code (0)

등록된 구현이 없습니다.

Tasks

Computational EfficiencyContrastive LearningData AugmentationRepresentation Learning

Methods 이 논문이 사용한 방법론

BYOL 설명 없음

Similar Papers 제목 키워드 기반

Understanding Dimensional Collapse in Contrastive Self-supervised Learning

2021-10-18 · ICLR 2022 4 · Li Jing, Pascal Vincent, Yann Lecun, Yuandong Tian

Self-supervised visual representation learning aims to learn useful representations without relying on human annotations. Joint embedding approach bases on maximizing the agreement between embedding vectors from differen…

Contrastive LearningLearning TheoryRepresentation LearningSelf-Supervised Learning

No Gaussian Required: Contrastive Inverse Dynamics for JEPA World Models

2026-08-18 · Jack Boylan, Chris Hokamp arxiv

Joint-Embedding Predictive Architectures (JEPAs) learn world models by predicting future embeddings, but the objective admits a trivial solution of a constant encoder, so every practical system adds an anti-collapse mech…

Relaxed Contrastive Learning for Federated Learning

2024-01-10 · CVPR 2024 1 · Seonguk Seo, Jinkyu Kim, Geeho Kim, Bohyung Han

We propose a novel contrastive learning framework to effectively address the challenges of data heterogeneity in federated learning. We first analyze the inconsistency of gradient updates across clients during local trai…

Contrastive LearningFederated Learning

Probabilistic Skip Connections for Deterministic Uncertainty Quantification in Deep Neural Networks

2025-01-08 · Felix Jimenez, Matthias Katzfuss

Deterministic uncertainty quantification (UQ) in deep learning aims to estimate uncertainty with a single pass through a network by leveraging outputs from the network's feature extractor. Existing methods require that t…

Out of Distribution (OOD) DetectionUncertainty Quantification

ContraNorm: A Contrastive Learning Perspective on Oversmoothing and Beyond

2023-03-12 · Xiaojun Guo, Yifei Wang, Tianqi Du, Yisen Wang

Oversmoothing is a common phenomenon in a wide range of Graph Neural Networks (GNNs) and Transformers, where performance worsens as the number of layers increases. Instead of characterizing oversmoothing from the view of…

Contrastive Learning