UCLA

Existing methods for graph change point detection focus on binary graphs, limiting their ability to capture the richer information available in weighted graphs. This work introduces a novel algorithm for detecting change points in weighted graphs by combining generative modeling with Group-Fused Lasso. Graphs are represented through low-dimensionallatent vectors, with neural networks modeling their underlying distributions. Specifically, change points are identified as shifts n the distributions of latent vectors, with GroupFused Lasso optimized using the Alternating Direction Method of Multipliers (ADMM). Simulation experiments and real-world applications demonstrate the effectiveness of the proposed method, highlighting the advantages of incorporating edge weights for a deeper understanding of temporal graph dynamics.