Efficient and portable GEMM-based convolution operators for deep neural network training on multicore processors | Publicación