Summary of Views Can Be Deceiving: Improved Ssl Through Feature Space Augmentation, by Kimia Hamidieh et al.
Views Can Be Deceiving: Improved SSL Through Feature Space Augmentation
by Kimia Hamidieh, Haoran Zhang, Swami Sankaranarayanan, Marzyeh Ghassemi
First submitted to arxiv on: 28 May 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: 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 This paper investigates how Self-Supervised Learning (SSL) methods in visual representation learning rely on spurious features for prediction. The authors demonstrate that commonly used augmentations in SSL can introduce undesired invariances in the image space, leading to suboptimal performance on minority subgroups. They also show that classical approaches to combating spurious correlations do not consistently lead to invariant representations. To address this issue, the authors propose a novel method called LateTVG, which regularizes later layers of the encoder via pruning to remove spurious information from the representations during pre-training. The proposed method outperforms baselines on several benchmarks without requiring group or label information during SSL. This work highlights the importance of understanding and addressing the impact of spurious features in SSL for visual representation learning. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper looks at how a type of machine learning called Self-Supervised Learning (SSL) works. The authors found that when using certain techniques to train this method, it can actually make things worse by introducing unwanted patterns in the images. They also tried some common ways to fix this problem and found that they didn’t work well. To solve this issue, the researchers came up with a new approach called LateTVG. It helps remove these unwanted patterns from the training data before using it for real-world applications. This method was shown to perform better than other methods on several tests without needing any extra information. |
Keywords
» Artificial intelligence » Encoder » Machine learning » Pruning » Representation learning » Self supervised
