Summary of Preventing Collapse in Contrastive Learning with Orthonormal Prototypes (clop), by Huanran Li et al.
Preventing Collapse in Contrastive Learning with Orthonormal Prototypes (CLOP)
by Huanran Li, Manh Nguyen, Daniel Pimentel-Alarcón
First submitted to arxiv on: 27 Mar 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI)
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 has emerged as a powerful method in deep learning, excelling at learning effective representations through contrasting samples from different distributions. However, neural collapse poses a significant challenge, especially in semi-supervised and self-supervised setups. This paper theoretically analyzes the effect of large learning rates on contrastive losses that solely rely on the cosine similarity metric, deriving a theoretical bound to mitigate this collapse. Building on these insights, the authors propose CLOP, a novel semi-supervised loss function designed to prevent neural collapse by promoting the formation of orthogonal linear subspaces among class embeddings. Unlike prior approaches that enforce a simplex ETF structure, CLOP focuses on subspace separation, leading to more distinguishable embeddings. Extensive experiments on real and synthetic datasets demonstrate that CLOP enhances performance, providing greater stability across different learning rates and batch sizes. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about how we can make machines learn better by comparing samples from different groups. Right now, there’s a problem called “neural collapse” where the machine gets stuck in a simpler way of thinking. The authors figured out why this happens and came up with a new way to stop it. Their method, called CLOP, helps the machine learn more effectively by making sure the different groups stay distinct. They tested their idea on many datasets and found that it works better than other methods. |
Keywords
* Artificial intelligence * Cosine similarity * Deep learning * Loss function * Self supervised * Semi supervised
