Summary of Libragrad: Balancing Gradient Flow For Universally Better Vision Transformer Attributions, by Faridoun Mehri (1) et al.
LibraGrad: Balancing Gradient Flow for Universally Better Vision Transformer Attributions
by Faridoun Mehri, Mahdieh Soleymani Baghshah, Mohammad Taher Pilehvar
First submitted to arxiv on: 24 Nov 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI); 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 paper addresses the issue of gradient-based explanations struggling with Transformers, specifically identifying gradient flow imbalances that violate FullGrad-completeness. To correct this, LibraGrad is introduced as a post-hoc approach that prunes and scales backward paths without changing the forward pass or adding computational overhead. Evaluations across 8 architectures, 4 model sizes, and 4 datasets demonstrate LibraGrad’s universality in enhancing gradient-based methods, outperforming existing white-box methods, including Transformer-specific approaches. This is achieved through three metric families: Faithfulness, Completeness Error, and Segmentation AP. The paper also presents superior qualitative results through text-prompted region highlighting on CLIP models and class discrimination between co-occurring animals on ImageNet-finetuned models. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper explores why some explanations (like gradients) have trouble understanding Transformers. It finds that the way gradients flow is imbalanced, which makes them less accurate. To fix this, the researchers created a new method called LibraGrad, which corrects these imbalances without changing how the model works or adding extra calculations. They tested LibraGrad on many different models and datasets and found it worked better than other methods. The paper also shows that LibraGrad can help explain complex data like images and text. |
Keywords
» Artificial intelligence » Transformer
