Method for updating artificial neural network

Abstract

According to an aspect, the present disclosure proposes a method for updating an artificial neural network, the artificial neural network comprising initial weights stored in a memory at least in an integer format, the method comprising: a processing unit determining an error gradient at an output of a layer of the neural network; the processing unit retrieves the initial weight from the memory; the processing unit updates the initial weights, including: for each initial weight, performing a first calculation of the corrected weight in an integer format of the initial weight; the processing unit replaces the value of the initial weight stored in the memory with the value of the corrected weight.

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS [0002] This application claims priority to French Patent Application No. 2012081, filed on November 24, 2020, which is hereby incorporated by reference in its entirety. technical field [0003] The present disclosure relates generally to artificial neural networks, and in particular embodiments, to updating weights in layers of an artificial neural network. Background technique [0004] An artificial neural network is used to perform a given function while it is running. For example, one function of a neural network could be classification. Another function may consist essentially in generating a signal from a received input signal. [0005] Artificial neural networks usually include a continuum of neuron layers. Each layer receives weighted data at its input and outputs data after being processed by the layer's neuron activation function. The output data is then passed to the next layer in the neural network. [0006] W...

AI Technical Summary

Problems solved by technology

However, hardware platforms typically have a limited amount of power

Images