Low-memory GEMM-based convolution algorithms for deep neural networks

Re-assign community