Anomaly Detection in Dynamic Graphs via Adversarial Autoencoder
Keywords: anomaly detection, dynamic graph, graph learning, deep learning, autoencoder
TL;DR: AAEDY is a semi-supervised dynamic graph anomaly detection framework that introduces adversarial based on autoencoders, detects anomalies by comparing raw edges, and is tested on six datasets to verify its effectiveness.
Abstract: Anomaly detection in dynamic graphs is a very important task that has attracted a lot of attention. Many dynamic graph anomaly detection methods are already available, but most of these efforts are based on supervised learning. In the real world, however, it is often difficult to collect large amounts of labelled anomaly data, which is not conducive to the training of these supervised methods and severely reduces their ability to be applied in different dynamic graph anomaly detection scenarios. A novel semi-supervised anomaly detection framework \textbf{AAEDY} for the detection of anomalous edges in dynamic graphs is presented in this paper, which improves reconstruction by combining adversarial based on autoencoder, and discriminates whether an edge is anomalous by comparing the original edge to the reconstructed edge in low-dimensional space. Extensive experiments have been carried out on six real-world datasets, and the experimental results show that \textbf{AAEDY} can outperform the state-of-the-art competitors in anomaly detection significantly.
Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation learning
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 10468
Loading