Summary of Gdda: Semantic Ood Detection on Graphs Under Covariate Shift Via Score-based Diffusion Models, by Zhixia He and Chen Zhao and Minglai Shao and Yujie Lin and Dong Li and Qin Tian
GDDA: Semantic OOD Detection on Graphs under Covariate Shift via Score-Based Diffusion Models
by Zhixia He, Chen Zhao, Minglai Shao, Yujie Lin, Dong Li, Qin Tian
First submitted to arxiv on: 23 Oct 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 a significant challenge in Out-of-distribution (OOD) detection for Graph Neural Networks (GNNs), focusing on open-world scenarios with varying distribution shifts. Most existing methods primarily focus on identifying instances caused by either semantic or covariate shifts, leaving the simultaneous occurrence of both types under-explored. This work introduces a novel challenge: graph-level semantic OOD detection under covariate shift, where variations between training and test domains result from concurrent presence of both shifts. To tackle this, the paper proposes GDDA, a two-phase framework that disentangles graph representations into domain-invariant semantic factors and domain-specific style factors. The first phase focuses on disentangling these factors, while the second phase employs a novel distribution-shift-controlled score-based generative diffusion model to generate latent factors outside the training semantic and style spaces. Additionally, pseudo-InD and pseudo-OOD graph representations are used to enhance the effectiveness of an energy-based semantic OOD detector. The approach outperforms state-of-the-art baselines on three benchmark datasets. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper solves a problem with Graph Neural Networks (GNNs) that helps them detect when data is not from the usual type they’re used to. Usually, GNNs are good at detecting changes in data, but they struggle when there’s both a change in what kind of data it is and how the data looks. The paper introduces a new challenge for GNNs: recognizing this type of change on graphs. To solve this, the researchers created a two-part plan called GDDA. The first part separates the graph into different parts that are important for understanding what’s happening in the data. The second part uses a special model to generate new patterns that are not like anything the GNN has seen before. This helps the GNN detect when the data is unusual and doesn’t belong. The results show that this approach works better than other methods. |
Keywords
* Artificial intelligence * Diffusion model * Gnn
