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

Standard deep learning algorithms are implemented using floating-point realnumbers. This presents an obstacle for implementing them on low-end deviceswhich may not have dedicated floating-point units (FPUs). As a result,researchers in tinyML have considered machine learning algorithms that cantrain and run a deep neural network (DNN) on a low-end device using integeroperations only. In this paper we propose PocketNN, a light and self-containedproof-of-concept framework in pure C++ for the training and inference of DNNsusing only integers. Unlike other approaches, PocketNN directly operates onintegers without requiring any explicit quantization algorithms or customizedfixed-point formats. This was made possible by pocket activations, which are afamily of activation functions devised for integer-only DNNs, and an emergingDNN training algorithm called direct feedback alignment (DFA). Unlike thestandard backpropagation (BP), DFA trains each layer independently, thusavoiding integer overflow which is a key problem when using BP withinteger-only operations. We used PocketNN to train some DNNs on two well-knowndatasets, MNIST and Fashion-MNIST. Our experiments show that the DNNs trainedwith our PocketNN achieved 96.98% and 87.7% accuracies on MNIST andFashion-MNIST datasets, respectively. The accuracies are very close to theequivalent DNNs trained using BP with floating-point real number operations,such that accuracy degradations were just 1.02%p and 2.09%p, respectively.Finally, our PocketNN has high compatibility and portability for low-enddevices as it is open source and implemented in pure C++ without anydependencies.

PocketNN: Integer-only Training and Inference of Neural Networks via Direct Feedback Alignment and Pocket Activations in Pure C++