Summary of On Discprecncies Between Perturbation Evaluations Of Graph Neural Network Attributions, by Razieh Rezaei et al.
On Discprecncies between Perturbation Evaluations of Graph Neural Network Attributions
by Razieh Rezaei, Alireza Dizaji, Ashkan Khakzar, Anees Kazi, Nassir Navab, Daniel Rueckert
First submitted to arxiv on: 1 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 Graph neural networks (GNNs) are increasingly being applied to model relationships between features in graph data. To understand the relationships between nodes in graphs, researchers have developed various attribution methods to explain the outputs of GNNs. However, existing attribution methods differ significantly, and it is unclear which one to trust. To address this issue, researchers have introduced evaluation experiments that assess these attribution methods from different perspectives. This work evaluates attribution methods from a new perspective: retraining. The core idea is to retrain the network on important relationships as identified by the attributions and evaluate how networks generalize based on these relationships. By reformulating the retraining framework and proposing guidelines for correct analysis, this study assesses four state-of-the-art GNN attribution methods and five synthetic and real-world graph classification datasets. The results show that attributions perform variably depending on the dataset and network, with some performing similarly to arbitrary designations of edge importance. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Graphs are a way to represent relationships between things. Neural networks can learn from these graphs, but it’s hard to understand why they make certain predictions. To fix this problem, researchers have created different methods to explain how neural networks work on graphs. But these methods don’t all agree with each other, and it’s unclear which one is the best. This study looks at a new way to test these methods by retraining the network based on what they say are important relationships. The results show that different methods perform differently depending on the data and type of neural network used. |
Keywords
* Artificial intelligence * Classification * Gnn * Neural network
