A bundle-filter method for nonsmooth convex constrained optimization

摘要: For solving nonsmooth convex constrained optimization problems, we propose an algorithm which combines the ideas of proximal bundle methods with filter strategy for evaluating candidate points. The resulting inherits some attractive features from both approaches. On one hand, it allows effective control size quadratic programming subproblems via compression and aggregation techniques methods. other criterion accepting a point as new iterate is sometimes easier to satisfy than usual descent condition in Some encouraging preliminary computational results are also reported.