Efficient parallel algorithm for computing rough set approximation on GPU

Computation of rough set approximation (RSA) is a critical step for attribute reduction and knowledge acquisition in rough set theory. Continuously improving computation efficiency of RSA is very meaningful, because it can enhance user experience of existing applications. Furthermore, it is helpful to apply rough sets to some fields with high performance requirement. Graphics processing unit (GPU) has gained a lot of attention from scientific communities for its applicability in high-performance computing. Different from existing works, this paper tries to apply GPU to accelerate a state-of-the-art serial algorithm of RSA computation, which is based on radix sorting. Three key steps of the serial algorithm are parallel designed, including object sorting, computation of equivalence classes, and computation of RSA. The experimental results show that the parallel method can accelerate the computation process efficiently.