Summary of Decompose-and-compose: a Compositional Approach to Mitigating Spurious Correlation, by Fahimeh Hosseini Noohdani et al.
Decompose-and-Compose: A Compositional Approach to Mitigating Spurious Correlation
by Fahimeh Hosseini Noohdani, Parsa Hosseini, Aryan Yazdan Parast, Hamidreza Yaghoubi Araghi, Mahdieh Soleymani Baghshah
First submitted to arxiv on: 29 Feb 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 addresses the issue of distribution shift in image classification, where standard Empirical Risk Minimization (ERM) training performs poorly on out-of-distribution samples. The problem arises from the compositional nature of images, which can lead to spurious correlations with the label. To improve robustness to correlation shift, the authors propose Decompose-and-Compose (DaC), a compositional approach that combines elements of images. DaC is shown to improve model performance on out-of-distribution samples by identifying causal components and intervening on images. The method also has high interpretability and can be used with group-balancing methods without requiring group labels or information about spurious features during training. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper helps us understand how computers learn from pictures, but it’s not just about that! Imagine you’re trying to teach a computer to recognize different breeds of dogs. But what if the pictures are taken in different lighting conditions or with different camera angles? The computer would have trouble recognizing the right breed because it wasn’t trained on those specific scenarios. That’s kind of like what happens when we try to use a model that was trained on one type of picture for another type of picture. This paper shows us how to fix this problem by breaking down the pictures into smaller parts, or “compositional approach”, which helps the computer learn more about what makes a breed of dog look like it does. |
Keywords
* Artificial intelligence * Image classification
