Suboptimality bounds for trace-bounded SDPs enable a faster and scalable low-rank SDP solver SDPLR+
Researchers have developed a new semidefinite programming (SDP) solver called SDPLR+ that significantly improves scalability and speed for large-scale problems. This solver optimizes over low-rank factorizations, reducing storage costs compared to traditional methods. SDPLR+ dynamically adjusts the rank during optimization, enabling early termination and faster computation, and has demonstrated effectiveness on problems with up to million-by-million decision variables. AI
IMPACT Enhances computational capabilities for machine learning and data science problems involving large-scale optimization.