Summary of Probabilistic Contrastive Learning For Long-tailed Visual Recognition, by Chaoqun Du et al.
Probabilistic Contrastive Learning for Long-Tailed Visual Recognition
by Chaoqun Du, Yulin Wang, Shiji Song, Gao Huang
First submitted to arxiv on: 11 Mar 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Computer Vision and Pattern Recognition (cs.CV)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The proposed ProCo learning algorithm aims to alleviate the issue of class imbalance in supervised contrastive learning. By estimating the data distribution of samples from each class in the feature space and sampling contrastive pairs accordingly, ProCo addresses the challenge of requiring sufficiently large batches of training data. The authors introduce a simple assumption that normalized features follow a mixture of von Mises-Fisher (vMF) distributions on unit space, which allows for efficient estimation of distribution parameters using only the first sample moment. This leads to a closed-form expression of the expected contrastive loss for optimization. The ProCo algorithm is particularly effective in alleviating the data imbalance issue. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary ProCo is an innovative approach that helps machines learn from imbalanced data. Normally, when there’s too much more of one type than another, it’s hard to train models accurately. ProCo solves this problem by cleverly estimating the distribution of each class and sampling pairs in a way that makes sense for these types of datasets. It does this by assuming that features are spread out following certain patterns, which makes it easy to estimate parameters. This allows for efficient optimization and better performance on imbalanced data. |
Keywords
* Artificial intelligence * Contrastive loss * Optimization * Supervised
