Hiroshima University, Graduate School of advanced science Sciences of Nakano 浩嗣 Professor of the research team and NTT Data is a 8 month 17 days, the combination of optimization problem the solution high speed to explore a new calculation method”adaptive・bulk・search”to be completed.
The new method, a plurality of algorithms using the”quadratic unconstrained binary optimization (QUBO) problem”and a large amount of from the solution with the lowest cost as the optimal solution to the multiple GPUS for efficient parallel explore that. Machine number of units calculated in proportion to the speed you can speed up the features of the NVIDIA GPU to 4 groups with a compute server per second in 1 trillion over the solution to explore the computational speed is achieved. In the experiment, the maximum cut problem,The Traveling Salesman Problem,random problems that this method of Speed showed that,at present 3 million 2768 variable QUBO problem to treat even that.
Adaptive・bulk・search operation of the image (source:NTT Data)
Detailed results the same day, from being held an International Conference”International Conference on Parallel Processing”in Hiroshima Univ of 安戸 colleagues compared to the last adjunct assistant Professor released. NTT DATA,this solution leverages distribution and Finance, manufacturing issues, and artificial intelligence/machine learning for high performance working on reducing that.