Summary of Contrastive Learning with Negative Sampling Correction, by Lu Wang et al.
Contrastive Learning with Negative Sampling Correction
by Lu Wang, Chao Du, Pu Zhao, Chuan Luo, Zhangchi Zhu, Bo Qiao, Wei Zhang, Qingwei Lin, Saravan Rajmohan, Dongmei Zhang, Qi Zhang
First submitted to arxiv on: 13 Jan 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: None
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 Contrastive learning (CL) has been a powerful tool for self-supervised representation learning, relying on multiple negative pairs to contrast against each positive pair. However, existing works have often overlooked the negative sampling process, which can lead to biased losses and performance degradation due to polluted negative samples. To address this issue, we propose Positive-Unlabeled Contrastive Learning (PUCL), a novel method that treats generated negative samples as unlabeled data and corrects bias in contrastive loss using information from positive samples. PUCL outperforms state-of-the-art methods on various image and graph classification tasks. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Contrastive learning is a way to learn without labels by comparing similar and different things. It’s like looking at two pictures and saying, “These are the same!” or “These are very different!” But sometimes, the program making these comparisons gets it wrong and says that things are more alike than they really are. This can make the program not work as well. To fix this, we came up with a new way to do contrastive learning called Positive-Unlabeled Contrastive Learning (PUCL). PUCL is like a referee who makes sure the program isn’t making mistakes by comparing positive things to negative things in a special way. This helps the program learn better and work better on things like recognizing pictures or understanding graphs. |
Keywords
* Artificial intelligence * Classification * Contrastive loss * Representation learning * Self supervised
