Convergence-Aware Neural Network Training

Training a deep neural network (DNN) is expensive, requiring a large amount of computation time. While the training overhead is high, not all computation in DNN training is equal. Some parameters converge faster and thus their gradient computation may contribute little to the parameter update; in near-stationary points a subset of parameters may change very little. In this paper we exploit the parameter convergence to optimize gradient computation in DNN training. We design a light-weight monitoring technique to track the parameter convergence; we prune the gradient computation stochastically for a group of semantically related parameters, exploiting their convergence correlations. These techniques are efficiently implemented in existing GPU kernels. In our evaluation the optimization techniques substantially and robustly improve the training throughput for four DNN models on three public datasets.