Summary of Understanding and Mitigating Human-labelling Errors in Supervised Contrastive Learning, by Zijun Long and Lipeng Zhuang and George Killick and Richard Mccreadie and Gerardo Aragon Camarasa and Paul Henderson
Understanding and Mitigating Human-Labelling Errors in Supervised Contrastive Learning
by Zijun Long, Lipeng Zhuang, George Killick, Richard McCreadie, Gerardo Aragon Camarasa, Paul Henderson
First submitted to arxiv on: 10 Mar 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
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 paper investigates the impact of human-annotated mislabelled examples on Supervised Contrastive Learning (SCL) models. Unlike synthetic label errors, human mislabelling poses unique challenges in SCL, affecting the learning process in most cases when it occurs as false positive samples. Existing noise-mitigating methods, which primarily focus on high synthetic noise rates, often underperform on common image datasets due to overfitting. To address this issue, the authors propose a novel SCL objective with robustness to human-labelling errors (SCL-RHE), designed to mitigate the effects of real-world mislabelled examples. Experiments show that SCL-RHE consistently outperforms state-of-the-art representation learning and noise-mitigating methods across various vision benchmarks, offering improved resilience against human-labelling errors. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper looks at how humans making mistakes when labelling pictures affects a type of AI called Supervised Contrastive Learning (SCL). They find that this kind of mistake is different from the mistakes made by computers generating fake labels. This means SCL needs to be designed differently to handle these human mistakes. The authors come up with a new way for SCL to work better when there are mistakes in the labels, and they test it on lots of pictures. Their method does much better than other methods at learning from pictures. |
Keywords
* Artificial intelligence * Overfitting * Representation learning * Supervised
