■Title: Challenges to solve previously unsolved instances by using a large-scale parallel MIP solver
■Venue: W1-D-710/Live streaming with Zoom
■Speaker: SHINANO, Yuji (Zuse Institute Berlin)
■Attendance: 16(Students:4; Staff:6; Others:6)

Dr. Shinano introduced his researches to solve open problems in Mixed-Integer Programming (MIP) by combining latest solvers with large-scale parallel computing. He explained the development of “UG,” which is a framework designed to execute solvers such as SCIP in large-scale parallel computing. He also explained that UG uses techniques including racing ramp-up, dynamic load balancing, and checkpoint/restart, and introduced that its application has successfully obtained optimal solutions — for the first time — for more than 20 benchmark problems in the MIPLIB library.
Furthermore, after addressing specific to parallelization — such as changes in problem structure caused by preprocessing — Dr. Shinano reported on the development of a new method for sharing solutions and information across multiple solvers (including Gurobi and Xpress) and shown some results on applications that could not be handled by presence methods.