High Performance Relevance Vector Machine on GPUs