Efficient Stochastic Inference of Bitwise Deep Neural Networks
This work addresses the need for efficient embedded systems by enhancing bitwise neural networks, though it is incremental as it builds on existing bitwise training methods.
The paper tackles the problem of improving classification accuracy in bitwise neural networks by using ensemble decisions from multiple stochastically sampled models, achieving a 5.81% best classification error on the CIFAR-10 test set and surpassing high-precision base models.
Recently published methods enable training of bitwise neural networks which allow reduced representation of down to a single bit per weight. We present a method that exploits ensemble decisions based on multiple stochastically sampled network models to increase performance figures of bitwise neural networks in terms of classification accuracy at inference. Our experiments with the CIFAR-10 and GTSRB datasets show that the performance of such network ensembles surpasses the performance of the high-precision base model. With this technique we achieve 5.81% best classification error on CIFAR-10 test set using bitwise networks. Concerning inference on embedded systems we evaluate these bitwise networks using a hardware efficient stochastic rounding procedure. Our work contributes to efficient embedded bitwise neural networks.