Summary of Rethinking Self-distillation: Label Averaging and Enhanced Soft Label Refinement with Partial Labels, by Hyeonsu Jeong and Hye Won Chung
Rethinking Self-Distillation: Label Averaging and Enhanced Soft Label Refinement with Partial Labels
by Hyeonsu Jeong, Hye Won Chung
First submitted to arxiv on: 16 Feb 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Machine Learning (stat.ML)
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 research paper investigates self-distillation mechanisms in multi-class classification, specifically in linear probing with fixed feature extractors. The authors provide a theoretical analysis revealing that multi-round self-distillation performs label averaging among instances with high feature correlations, governed by the eigenvectors of the Gram matrix. This process leads to clustered predictions and improved generalization, mitigating the impact of label noise. The paper also introduces a novel single-round self-distillation method using refined partial labels from the teacher’s top two softmax outputs, achieving comparable or superior performance in high-noise scenarios while reducing computational cost. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper looks at how machines can improve their decision-making when they’re not perfect and there are mistakes. It finds that repeating a process to get better and better makes sense. This helps the machine make more accurate decisions, especially when there’s a lot of noise or errors in the information it gets. The researchers also come up with a new way for machines to learn that’s faster and just as good. |
Keywords
* Artificial intelligence * Classification * Distillation * Generalization * Softmax
