Summary of Understanding the Detrimental Class-level Effects Of Data Augmentation, by Polina Kirichenko et al.
Understanding the Detrimental Class-level Effects of Data Augmentation
by Polina Kirichenko, Mark Ibrahim, Randall Balestriero, Diane Bouchacourt, Ramakrishna Vedantam, Hamed Firooz, Andrew Gordon Wilson
First submitted to arxiv on: 7 Dec 2023
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 The proposed framework sheds light on how data augmentation (DA) interacts with class-level learning dynamics in image classification tasks, particularly when dealing with highly ambiguous or co-occurring classes. By leveraging higher-quality multi-label annotations on ImageNet, researchers categorize the affected classes and identify sources of accuracy degradation beyond just DA’s bias towards one closely related class. The findings suggest that simple class-conditional augmentation strategies can improve performance on negatively impacted classes. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Data augmentation helps machines recognize images better by adding fake data to training sets. But sometimes this boost in accuracy comes at a cost: certain types of images are hurt more than others. A new study explains why this happens and how it can be fixed. By looking at pictures more closely, researchers found that many problems come from ambiguous or related images. To fix these issues, they developed special ways to add fake data for each type of image. |
Keywords
* Artificial intelligence * Data augmentation * Image classification
