Summary of Debiased Sample Selection For Combating Noisy Labels, by Qi Wei et al.
Debiased Sample Selection for Combating Noisy Labels
by Qi Wei, Lei Feng, Haobo Wang, Bo An
First submitted to arxiv on: 24 Jan 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 The paper addresses the issue of learning with noisy labels, which aims to ensure model generalization despite a label-corrupted training set. A sample selection strategy can achieve promising performance by selecting a reliable subset for model training. However, existing methods suffer from both data and training bias, represented as imbalanced selected sets and accumulation errors in practice. To address this limitation, the authors propose a noIse-Tolerant Expert Model (ITEM) for debiased learning in sample selection. ITEM integrates multiple experts to mitigate training bias and proposes a mixed sampling strategy to mitigate data bias. The model is trained on a mixture of two class-discriminative mini-batches to avoid sparse representations caused by sampling strategies. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper looks at how to learn with noisy labels, which is important for machine learning models to work well in real-world situations. When the training data has incorrect labels, it can be hard for the model to generalize well. The authors find that existing methods don’t do a good job of handling both types of bias: data bias and training bias. They propose a new approach called ITEM, which combines multiple experts to reduce bias during training and uses a special way of sampling the data to reduce bias in the data itself. |
Keywords
* Artificial intelligence * Generalization * Machine learning
