Training Recurrent Neural Networks against Noisy Computations during Inference
This addresses power efficiency issues in embedded systems like speech recognition by enabling robust neural networks for noisy analog hardware, though it is incremental as it adapts existing training methods.
The paper tackles the problem of recurrent neural networks being vulnerable to noisy computations during inference, which arises from using low-power analog neuromorphic circuits, and shows that Deep Noise Injection training makes the networks more robust across a wide range of noise powers, with some networks even improving performance in noise-free conditions.
We explore the robustness of recurrent neural networks when the computations within the network are noisy. One of the motivations for looking into this problem is to reduce the high power cost of conventional computing of neural network operations through the use of analog neuromorphic circuits. Traditional GPU/CPU-centered deep learning architectures exhibit bottlenecks in power-restricted applications, such as speech recognition in embedded systems. The use of specialized neuromorphic circuits, where analog signals passed through memory-cell arrays are sensed to accomplish matrix-vector multiplications, promises large power savings and speed gains but brings with it the problems of limited precision of computations and unavoidable analog noise. In this paper we propose a method, called {\em Deep Noise Injection training}, to train RNNs to obtain a set of weights/biases that is much more robust against noisy computation during inference. We explore several RNN architectures, such as vanilla RNN and long-short-term memories (LSTM), and show that after convergence of Deep Noise Injection training the set of trained weights/biases has more consistent performance over a wide range of noise powers entering the network during inference. Surprisingly, we find that Deep Noise Injection training improves overall performance of some networks even for numerically accurate inference.