Reinforced Causal Explainer for Graph Neural Networks, TPAMI2022

Introduction of RGCNExplainer, an explainability approach for Relational Graph Convolutional Neural Networks.

EDGE, "Evaluation of Diverse Knowledge Graph Explanations", is a framework to benchmark diverse explanations (e.g., subgraph vs logical) for node classification in knowledge graphs.

Relational Deep Learning and Explainability of Graph Neural Network

Fraud Detection System using Graph Neural Networks (AD-RL-GNN) to identify complex fraud patterns. Features: 22.7% G-Means improvement over baseline, <28ms real-time latency, Adaptive Majority Downsampling (MCD) for 28:1 class imbalance, and a scalable MLOps pipeline (FastAPI, Redis, Docker).