Year: 2017
Type: conference-paper
Abstract: Due to the recent congruence of big data and compute acceleration, the field of machine learning (ML) has made significant advancements, driven primarily by deep neural networks trained with the stochastic gradient descent algorithm (DNN). Error rates on challenging learning benchmarks have fallen precipitously. The algorithmic success of DNNs combined with the significant architectural mismatch between modern processors and neural architectures has resulted in a race to build DNN processors. The mismatch is demonstrated by analyzing the dynamic power consumption due to capacitive loss associated with synaptic access:
Authors: Alex Nugent, Tim Molter
Language: English
Subfield: Electrical and Electronic Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: gold
OpenAlex ID: W2802783863