English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Tools annual

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

We explore the robustness of recurrent neural networks when the computationswithin the network are noisy. One of the motivations for looking into thisproblem is to reduce the high power cost of conventional computing of neuralnetwork operations through the use of analog neuromorphic circuits. TraditionalGPU/CPU-centered deep learning architectures exhibit bottlenecks inpower-restricted applications, such as speech recognition in embedded systems.The use of specialized neuromorphic circuits, where analog signals passedthrough memory-cell arrays are sensed to accomplish matrix-vectormultiplications, promises large power savings and speed gains but brings withit the problems of limited precision of computations and unavoidable analognoise.  In this paper we propose a method, called {\em Deep Noise Injectiontraining}, to train RNNs to obtain a set of weights/biases that is much morerobust against noisy computation during inference. We explore several RNNarchitectures, such as vanilla RNN and long-short-term memories (LSTM), andshow that after convergence of Deep Noise Injection training the set of trainedweights/biases has more consistent performance over a wide range of noisepowers entering the network during inference. Surprisingly, we find that DeepNoise Injection training improves overall performance of some networks even fornumerically accurate inference.

Training Recurrent Neural Networks against Noisy Computations during Inference