Deep learning for systemic financial risk: an Unsupervised Graph Autoencoder and a Spatio-Temporal GNN (GCN+LSTM) trained on 26 years of BIS banking network data (2000–2026) to autonomously detect crises and forecast cross-border exposure shifts — no manual features, no crisis labels.
deep-learning network-science pytorch lstm banking unsupervised-learning spatio-temporal gcn anomaly-detection early-warning-systems time-series-forecasting graph-neural-networks graph-representation-learning systemic-risk pytorch-geometric graph-autoencoder temporal-gnn bis-data financial-learning macroprudential-risk
-
Updated
Sep 7, 2026 - Jupyter Notebook