End-to-End Neural Network Compression via $\frac{{\ell }_1}{{\ell }_2}$ Regularized Latency Surrogates

Anshul Nasery, Hardik Shah, Arun Sai Suggala +1

Neural network (NN) compression via techniques such as pruning, quantization requires setting compression hyperparameters (e.g., number of channels to be pruned, bitwidths for quan…