A Directed-Evolution Method for Sparsification and Compression of Neural Networks with Application to Object Identification and Segmentation and considerations of optimal quantization using small number of bits.