Skip to main navigation Skip to search Skip to main content

Adaptive precision neural networks for image classification

  • Stony Brook University

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

3 Scopus citations

Abstract

We present a technique and algorithms to solve the following problem: Given both a Neural Network trained to classify a set of images, along with a set of floating-point hardware blocks (in reconfigurable logic), find the arrangement of blocks that achieves the best mix of precision, resources and speed with respect to a given cost function. We first illustrate the technique in detail by using a small example, then show that it may be used for a larger problem, bar code classification.

Original languageEnglish
Title of host publicationProceedings of the 2008 NASA/ESA Conference on Adaptive Hardware and Systems, AHS 2008
Pages244-251
Number of pages8
DOIs
StatePublished - 2008
Event2008 NASA/ESA Conference on Adaptive Hardware and Systems, AHS 2008 - Noordwijk, Netherlands
Duration: Jun 22 2008Jun 25 2008

Publication series

NameProceedings of the 2008 NASA/ESA Conference on Adaptive Hardware and Systems, AHS 2008

Conference

Conference2008 NASA/ESA Conference on Adaptive Hardware and Systems, AHS 2008
Country/TerritoryNetherlands
CityNoordwijk
Period06/22/0806/25/08

Fingerprint

Dive into the research topics of 'Adaptive precision neural networks for image classification'. Together they form a unique fingerprint.

Cite this