Summary of Rethinking the Impact Of Noisy Labels in Graph Classification: a Utility and Privacy Perspective, by De Li et al.
Rethinking the impact of noisy labels in graph classification: A utility and privacy perspective
by De Li, Xianxian Li, Zeming Gan, Qiyu Li, Bin Qu, Jinyan Wang
First submitted to arxiv on: 11 Jun 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 paper proposes a robust graph neural network approach for noisy labeled graph classification tasks. The authors find that existing methods degrade generalization performance and enhance membership inference attacks on graph data privacy when noisy labels are present. To address this, they propose a novel method that accurately filters noisy samples using high-confidence samples and the first feature principal component vector of each class. The approach also incorporates noise label correction guided by dual spatial information and supervised graph contrastive learning to enhance embedding quality and protect training graph data privacy. Experimental results on eight real graph classification datasets demonstrate a performance gain of up to 7.8% compared to state-of-the-art methods at 30% noisy labeling rate, while reducing the accuracy of privacy attacks to below 60%. The authors’ contributions have significant implications for both model utility and data privacy in graph classification tasks. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Imagine you’re trying to teach a computer to recognize patterns in graphs. But what if the training data has mistakes or “noise”? That’s exactly the problem this paper solves. The researchers found that current methods don’t work well when there are noisy labels. They created a new approach that can correct these errors and protect the privacy of the training data. In tests on eight real datasets, their method outperformed others by up to 7.8% at a certain level of noise, while also reducing the accuracy of attacks on the data’s privacy. |
Keywords
» Artificial intelligence » Classification » Embedding » Generalization » Graph neural network » Inference » Supervised
