Timezone: »

 
Spotlight
An Investigation into Whitening Loss for Self-supervised Learning
Xi Weng · Lei Huang · Lei Zhao · Rao Anwer · Salman Khan · Fahad Shahbaz Khan

Tue Dec 06 09:00 AM -- 11:00 AM (PST) @
in Lightning Talks 1B-3 »

A desirable objective in self-supervised learning (SSL) is to avoid feature collapse. Whitening loss guarantees collapse avoidance by minimizing the distance between embeddings of positive pairs under the conditioning that the embeddings from different views are whitened. In this paper, we propose a framework with an informative indicator to analyze whitening loss, which provides a clue to demystify several interesting phenomena as well as a pivoting point connecting to other SSL methods. We reveal that batch whitening (BW) based methods do not impose whitening constraints on the embedding, but they only require the embedding to be full-rank. This full-rank constraint is also sufficient to avoid dimensional collapse. Based on our analysis, we propose channel whitening with random group partition (CW-RGP), which exploits the advantages of BW-based methods in preventing collapse and avoids their disadvantages requiring large batch size. Experimental results on ImageNet classification and COCO object detection reveal that the proposed CW-RGP possesses a promising potential for learning good representations. The code is available at https://github.com/winci-ai/CW-RGP.

Keywords: Self-supervised learning · normalization · feature decorrelation · whitening
[ Slides] 

Author Information

Xi Weng (Beijing University of Aeronautics and Astronautics)
Lei Huang (Beihang University)

Lei Huang received his BSc and PhD degrees under supervision of Prof. Wei Li, respectively in 2010 and 2018, at the School of Computer Science and Engineering, Beihang University, China. From 2015 to 2016, he visited the Vision and Learning Lab, University of Michigan, Ann Arbor, as a joint PhD student supervised by Prof. Jia Deng. During 2018 to 2020, he was a research scientist in Inception Institute of Artificial Intelligence (IIAI), UAE. His current research mainly focuses on normalization techniques (involving methods, theories and applications) in training DNNs. He also has wide interests in deep learning theory (representation & optimization) and computers vision tasks. He serves as a reviewer for the top conferences and journals such as CVPR, ICML, ICCV, ECCV, NeurIPS, AAAI, JMLR, TPAMI, IJCV, TNNLS, etc.

Lei Zhao (TalkingData)
Rao Anwer (Mohamed bin Zayed University of Artificial Intelligence)
Salman Khan (MBZ University of AI)
Fahad Shahbaz Khan (Inception Institute of Artificial Intelligence)

Related Events (a corresponding poster, oral, or spotlight)

More from the Same Authors