Summary of Mitigating Noisy Supervision Using Synthetic Samples with Soft Labels, by Yangdi Lu et al.
Mitigating Noisy Supervision Using Synthetic Samples with Soft Labels
by Yangdi Lu, Wenbo He
First submitted to arxiv on: 22 Jun 2024
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 This paper investigates the challenges of training deep neural networks with noisy labels, a common issue in large-scale datasets. Noisy labels can lead to overfitting and poor generalization performance, making it crucial to develop effective strategies for mitigating their impact. The authors propose a framework that trains models with synthetic samples generated by aggregating original samples with their top-K nearest neighbors. This approach leverages the representation distributions in the early learning phase, where learned representations of images from the same category tend to congregate together. To enhance performance in the presence of extreme label noise, the authors estimate soft targets by gradually correcting noisy labels. Experimental results on two benchmarks (CIFAR-10 and CIFAR-100) and two large-scale real-world datasets (Clothing1M and Webvision) demonstrate that the proposed method outperforms state-of-the-art methods in terms of learned representation quality. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper looks at a big problem with training artificial intelligence models. Sometimes, the labels or answers we give to these models are wrong, which can make them not work well. The researchers found that if they mixed up old images with new ones that were similar, it helped the model learn better. They also came up with a way to fix noisy labels by slowly correcting mistakes. This method worked really well on big datasets and was even better than other methods. |
Keywords
» Artificial intelligence » Generalization » Overfitting
