CUDA를 활용한 계층적 군집화의 성능 향상

As the performance of graphics hardware rapidly increases, general-purpose computation on graphics processing units (GPGPUs) is emerging in various fields today. NVIDIA Corp. has introduced a GPGPU development environment called CUDA to facilitate GPGPU programming. In this paper, we propose a CUDA-based approach to speedup hierarchical clustering with single-link. The computation of pairwise distance matrices is accelerated by generating many threads. Our experimental results show that CUDA-based hierarchical clustering is approximately twice faster than traditional methods.