Special-Purpose Parallel Architectures for High-Performance Machine Learning