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 Plus annual

Presents the access of premium content as premium feature

Premium Content

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 Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

The power budget for embedded hardware implementations of Deep Learningalgorithms can be extremely tight. To address implementation challenges in suchdomains, new design paradigms, like Approximate Computing, have drawnsignificant attention. Approximate Computing exploits the innateerror-resilience of Deep Learning algorithms, a property that makes themamenable for deployment on low-power computing platforms. This paper describesan Approximate Computing design methodology, AX-DBN, for an architecturebelonging to the class of stochastic Deep Learning algorithms known as DeepBelief Networks (DBNs). Specifically, we consider procedures for efficientlyimplementing the Discriminative Deep Belief Network (DDBN), a stochastic neuralnetwork which is used for classification tasks, extending ApproximationComputing from the analysis of deterministic to stochastic neural networks. Forthe purpose of optimizing the DDBN for hardware implementations, we explore theuse of: (a)Limited precision of neurons and functional approximations ofactivation functions; (b) Criticality analysis to identify nodes in the networkwhich can operate at reduced precision while allowing the network to maintaintarget accuracy levels; and (c) A greedy search methodology with incrementalretraining to determine the optimal reduction in precision for all neurons tomaximize power savings. Using the AX-DBN methodology proposed in this paper, wepresent experimental results across several network architectures that showsignificant power savings under a user-specified accuracy loss constraint withrespect to ideal full precision implementations.

AX-DBN: An Approximate Computing Framework for the Design of Low-Power Discriminative Deep Belief Networks