A Nonmonotone Gradient-Based Algorithm for Symmetric Nonnegative Matrix Factorization and Graph Clustering
Researchers have developed SNMPBB, a novel nonmonotone projected Barzilai-Borwein algorithm for Symmetric Nonnegative Matrix Factorization (Symmetric NMF). This new method significantly improves convergence speed compared to existing projected gradient approaches for Symmetric NMF, achieving up to a six-fold speedup on synthetic data. The algorithm has been extended for graph clustering (Graph-SNMPBB) and large-scale problems with low-rank approximations (LAI-SNMPBB), demonstrating competitive accuracy and performance on real-world benchmarks and large matrices. AI
IMPACT Introduces a faster algorithm for matrix factorization, potentially improving performance in downstream machine learning and graph analysis tasks.