Pattern Decomposition with Complex Combinatorial Constraints: Application to Materials Discovery

摘要: Identifying important components or factors in large amounts of noisy data is a key problem machine learning and mining. Motivated by pattern decomposition materials discovery, aimed at discovering new for renewable energy, e.g. fuel solar cells, we introduce CombiFD, framework factor based that allows the incorporation a-priori knowledge as constraints, including complex combinatorial constraints. In addition, propose algorithm, called AMIQO, on solving sequence (mixed-integer) quadratic programs. Our approach considerably outperforms state art discovery problem, scaling to larger datasets recovering more precise physically meaningful decompositions. We also show effectiveness our enforcing background other application domains.