Continuous optimization plays a central role in modern machine learning. Every learning process requires solving a high-dimensional minimization problem. The faster we solve it, the fewer resources (e.g., energy, time, or money) we waste.
Current state-of-the-art algorithms include some heuristics and often require a lot of tuning. The theoretical justification of these heuristics and the developing of new, more efficient algorithms is the main focus of our group.
Current state-of-the-art algorithms include some heuristics and often require a lot of tuning. The theoretical justification of these heuristics and the developing of new, more efficient algorithms is the main focus of our group.
