Summary of Unsupervised Concept Discovery Mitigates Spurious Correlations, by Md Rifat Arefin et al.
Unsupervised Concept Discovery Mitigates Spurious Correlations
by Md Rifat Arefin, Yan Zhang, Aristide Baratin, Francesco Locatello, Irina Rish, Dianbo Liu, Kenji Kawaguchi
First submitted to arxiv on: 20 Feb 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Computer Vision and Pattern Recognition (cs.CV)
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 tackles the issue of models producing brittle predictions and unintended biases due to spurious correlations in training data. To address this challenge, researchers typically rely on prior knowledge and group annotation to remove these correlations. However, this approach may not be feasible in many applications. The authors propose a novel connection between unsupervised object-centric learning and mitigating spurious correlations. Instead of inferring subgroups with varying correlations with labels, the approach focuses on discovering concepts: discrete ideas shared across input samples. The authors introduce CoBalT, a concept balancing technique that effectively mitigates spurious correlations without requiring human labeling of subgroups. Experimental results demonstrate superior or competitive performance compared to state-of-the-art baselines on benchmark datasets for sub-population shifts. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper helps us make better predictions by avoiding mistakes caused by patterns in the data that are not really important. Right now, we often need a lot of human help and special knowledge to avoid these mistakes. The researchers found a new way to do this using object-centric learning, which is like learning about groups of things that share certain characteristics. They call their new method CoBalT, and it can fix the problems caused by spurious correlations without needing humans to label all the different groups. This means we can use CoBalT in situations where we don’t have a lot of human help or special knowledge. |
Keywords
* Artificial intelligence * Unsupervised
