Summary of Adaptive Group Robust Ensemble Knowledge Distillation, by Patrik Kenfack et al.
Adaptive Group Robust Ensemble Knowledge Distillation
by Patrik Kenfack, Ulrich Aïvodji, Samira Ebrahimi Kahou
First submitted to arxiv on: 22 Nov 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: None
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 study investigates knowledge distillation in neural networks, highlighting a significant performance disparity for underrepresented subgroups. The researchers found that traditional ensemble methods can actually worsen the issue when distilling knowledge from debiased teacher models. To address this problem, they propose Adaptive Group Robust Ensemble Knowledge Distillation (AGRE-KD), a simple strategy to ensure the student model receives beneficial knowledge for unknown underrepresented subgroups. The method uses an additional biased model to selectively choose teachers whose knowledge improves the worst-performing subgroups. Experimental results on several datasets demonstrate the superiority of this proposed ensemble distillation technique, even outperforming classic model ensembles based on majority voting. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This study looks at how neural networks learn and share information. It shows that when complex models teach simpler ones, it can actually make things worse for certain groups of people. The researchers developed a new way to help the student model learn from many teacher models, so it gets better information that helps everyone equally. They tested this method on several datasets and found it works really well, even beating traditional methods. |
Keywords
* Artificial intelligence * Distillation * Knowledge distillation * Student model
