Summary of Poly-view Contrastive Learning, by Amitis Shidani et al.
Poly-View Contrastive Learning
by Amitis Shidani, Devon Hjelm, Jason Ramapuram, Russ Webb, Eeshan Gunesh Dhekane, Dan Busbridge
First submitted to arxiv on: 8 Mar 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Information Theory (cs.IT); Machine Learning (stat.ML)
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 method extends contrastive learning by introducing poly-view tasks, which involve matching multiple related views. Building upon information maximization and sufficient statistics, novel representation learning objectives are derived. The paper demonstrates that with unlimited computation, it is beneficial to maximize the number of related views, while a fixed compute budget suggests decreasing unique samples and increasing view numbers. Specifically, poly-view contrastive models outperform SimCLR on ImageNet1k when trained for 128 epochs with a batch size of 256. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Contrastive learning usually matches similar views with many unrelated views. Researchers are now exploring what happens when there are more than two similar views. They came up with new ways to learn from these “poly-view” tasks using information and statistics. The study shows that if you had unlimited computing power, it would be best to focus on matching as many similar views as possible. However, with limited resources, it’s better to reduce the number of unique samples and increase the number of views for those samples. Interestingly, a poly-view contrastive model outperformed SimCLR on ImageNet1k when trained for 128 epochs with a batch size of 256. |
Keywords
* Artificial intelligence * Representation learning
