Mutual Conditional Probability for Self-Supervised Learning


Takumi Kobayashi (National Institute of Advanced Industrial Science and Technology)*
The 33rd British Machine Vision Conference

Abstract

Deep neural networks produce effective image feature representation through a supervised learning with plenty of annotation. To mitigate the label-hungry issue, self-supervised learning (SSL) works well for training the deep models without manual supervision. In SSL, pair-wise matching is widely applied to multi-view images via data augmentation techniques such as in a contrastive learning. In this work, we focus on a probabilistic distribution of the multi-view samples to effectively exploit the relationship among them which the pair-wise approaches hardly take into account. It leads to a novel loss based on mutual conditional probability through connecting SSL with mode seeking on the distribution. The method also exhibits connection to the other SSL methods. In the experiments on ImageNet classification tasks, the proposed method produces favorable performance in the framework of SSL.

Video



Citation

@inproceedings{Kobayashi_2022_BMVC,
author    = {Takumi Kobayashi},
title     = {Mutual Conditional Probability for Self-Supervised Learning},
booktitle = {33rd British Machine Vision Conference 2022, {BMVC} 2022, London, UK, November 21-24, 2022},
publisher = {{BMVA} Press},
year      = {2022},
url       = {https://bmvc2022.mpi-inf.mpg.de/1052.pdf}
}


Copyright © 2022 The British Machine Vision Association and Society for Pattern Recognition
The British Machine Vision Conference is organised by The British Machine Vision Association and Society for Pattern Recognition. The Association is a Company limited by guarantee, No.2543446, and a non-profit-making body, registered in England and Wales as Charity No.1002307 (Registered Office: Dept. of Computer Science, Durham University, South Road, Durham, DH1 3LE, UK).

Imprint | Data Protection