Raising the Bar in Graph-level Anomaly Detection

Abstract

Federated learning (FL) provides a privacy-preserving solution for distributed machine learning tasks. One challenging problem that severely damages the performance of FL models is the co-occurrence of data heterogeneity and long-tail distribution, which frequently appears in real FL applications. In this paper, we reveal an intriguing fact that the biased classifier is the primary factor leading to the poor performance of the global model. Motivated by the above finding, we propose a novel and privacy-preserving FL method for heterogeneous and long-tailed data via Classifier Re-training with Federated Features (CReFF). The classifier re-trained on federated features can produce comparable performance as the one re-trained on real data in a privacy-preserving manner without information leakage of local data or class distribution. Experiments on several benchmark datasets show that the proposed CReFF is an effective solution to obtain a promising FL model under heterogeneous and long-tailed data. Comparative results with the state-of-the-art FL methods also validate the superiority of CReFF. Our code is available at https://github.com/shangxinyi/CReFF-FL.

The danger of dichloromethane in the synthesis and application of imidazole-based dyes

“…[79] GNN [172], [173] Graph-level Homogeneous GA [174]-[179] GNN [72], [73], [180], [181] Heterogeneous GA [182] Dynamic GA [183], [184] GNN [185] TABLE 3: Summary of anomaly detection on graphs.…”

Graph Anomaly Detection With Graph Neural Networks: Current Status and Challenges

“…Other GNN-based model OCGNN [36] Targeting issue of GAE → GNN with hypersphere embedding space AAGNN [37] Targeting issue of GAE → GNN with hypersphere embedding space Meta-GDN [38] Hard work to label anomalies → meta-learning with auxiliary graphs Edge anomaly GCN-based GAE AANE [39] Noise or adversarial links → GAE with a loss for anomalous links eFraudCom [40] Fraud detection → heterogeneous graph and representative data sampling GCN alone SubGNN [41] Fraud detection → GIN and extracting and relabeling subgraphs Subgraph anomaly GCN alone GLocalKD [42] Graph-level anomalies → joint learning global & local normality GAT-based GAE HO-GAT [43] Abnormal subgraphs → hybrid-order attention with motif instances Graph-level anomaly GCN alone OCGIN [44] Graph-level anomaly detection → graph classification with GIN OCGTL [45] Hypersphere collapse → set of GNNs for embedding…”

“…To further improve the performance and overcome the hypersphere collapse where the deep one-class objective encourages all graph embeddings in the training data to concentrate within a hypersphere, Qiu et al [45] presented a new GNN-based approach, one-class graph transformation learning (OCGTL), which integrates advantages of deep one-class classification (deep OCC) [56], [44] and selfsupervised anomaly detection with learnable transformations [57]. Specifically, OCGTL consists of a set of GNNs to embed its input graphs into a latent space.…”

BOND: Benchmarking Unsupervised Outlier Node Detection on Static Attributed Graphs

“…In this work, we focus on the node-level detection with graphs due to its popularity, while there are more detection tasks at different levels of a graph. Recently, more graph OD algorithms are design for edge- [73], subgraph- [64], and graph-level [55,75] detection. A more holistic view of graph OD can include these emerging graph OD tasks.…”