Binary Matrix Factorisation via Column Generation

Identifying discrete patterns in binary data is an important dimensionality reduction tool machine learning and mining. In this paper, we consider the problem of low-rank matrix factorisation (BMF) under Boolean arithmetic. Due to hardness problem, most previous attempts rely on heuristic techniques. We formulate as a mixed integer linear program use large scale optimisation technique column generation solve it without need pattern Our approach focuses accuracy provision optimality guarantees. Experimental results real world datasets demonstrate that our proposed method effective at producing highly accurate factorisations improves previously available best known for 15 out 24 instances.