Edinburgh Research Explorer

Poster: Space and Time Optimal DNN Primitive Selection with Integer Linear Programming

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Related Edinburgh Organisations

Open Access permissions

Open

Documents

https://ieeexplore.ieee.org/document/8891651
Original languageEnglish
Title of host publication2019 28th International Conference on Parallel Architectures and Compilation Techniques (PACT)
PublisherIEEE
Pages488-489
Number of pages2
ISBN (Electronic)978-1-7281-3613-4
ISBN (Print)978-1-7281-3614-1
DOIs
Publication statusPublished - 7 Nov 2019
Event28th International Conference on Parallel Architectures and Compilation Techniques - Seattle, United States
Duration: 21 Sep 201925 Sep 2019
https://pactconf.org/

Publication series

Name
PublisherIEE
ISSN (Print)1089-795X
ISSN (Electronic)2641-7936

Conference

Conference28th International Conference on Parallel Architectures and Compilation Techniques
Abbreviated titlePACT 2019
CountryUnited States
CitySeattle
Period21/09/1925/09/19
Internet address

Abstract

Convolutional neural networks (CNNs) are used in many applications, from industrial robotics to biometric identification on mobile devices. But they can be too resourcehungry for mobile and embedded devices with tightly constrained memory and energy budgets. We propose an aheadof-time primitive selection for CNNs, based on integer linear programming (ILP). Under a tight memory budget, our ILP solver selects the optimal primitive for each layer such that the entire network is optimized for execution time subject to a memory budget, or vice versa. Our method yields significant speedup and memory reduction compared to existing methods.

    Research areas

  • neural network optimization, omputing operators, primitive selection, optimal convolutional layer

Event

28th International Conference on Parallel Architectures and Compilation Techniques

21/09/1925/09/19

Seattle, United States

Event: Conference

ID: 117347462