IORA take part in NUS National Day Observance Ceremony held on Thursday, 8 August 2019 at University Hall Lee Kong Chian Wing Atrium.

Short Description of Project:

The software SDPNAL+ is designed for solving semidefinite programming (SDP), an important subfield of mathematical optimization and its applications are growing rapidly. Many practical problems in operations research and machine learning can be modeled or approximated as SDP problems. Traditional optimization methods can only solve small and medium scale (say, matrix dimension is less than 2000 and the number of constraints is less than 5000) SDP. Fortunately, large-scale SDP can be solved efficiently by SDPNAL+ now. Numerical experiments in the paper and other benchmark tests show that SDPNAL+ is a state-of-the-art solver for large-scale SDP and it is the only viable software to solve many large-scale SDPs at present. The largest SDP problem that is solved has matrix dimension 9261 and the number of constraints more than 12 million, which boosts the solvable scale to thousands of times. This software, developed by IORA faculty Prof Toh Kim Chuan,  has won the 2018 Beale-Orchard-Hays Prize, the highest honor in the field of Computational Mathematical Optimization. In particular, the prize jury chair Dr. Michael Grant presented a concrete example shared by the nominator. It takes 122 hours for the traditional solver to solve a problem in a cluster with 56 cores CPU and 128 GPUs while SDPNAL+ solves it within 1.5 hours in a normal desktop PC. This new solver has many applications in practice. For instance, in a recent study, another IORA team of researchers have used this software to develop the backbone network structure to support bike rebalancing operations by volunteers in a system like the Bike Angel program in New York, and demonstrated big reduction in the number of redundant moves by volunteers in such a system (i.e. with much less incentives), but with essentially the same level of performance (number of rides supported).