2018 IEEE INTERNATIONAL CONFERENCE ON CONSUMER ELECTRONICS (ICCE)

Peter Corcoran

2018 IEEE INTERNATIONAL CONFERENCE ON CONSUMER ELECTRONICS (ICCE)

Electrical & Electronic Engineering

Research output: Chapter in Book or Conference Publication/Proceeding › Conference Publication › peer-review

Abstract

Transferring the success of deep learning models to consumer electronic devices requires the construction of deep learning models that are small enough to fit on resource constrained hardware. Since embedded and mobile devices lack the resources in terms of power consumption requirements, processing speed, and available memory of the latest GPU technology, it is desirable to create neural networks that are significantly smaller without sacrificing accuracy. A new technique for data augmentation called Smart Augmentation has recently been introduced that has been experimentally shown to be effective at this task. In this paper, we show how Smart Augmentation can be used to train models that are significantly smaller than their equivalently performant counterparts, and thus more viable for deployment on consumer devices.

Original language	English (Ireland)
Title of host publication	Learning data Augmentation for Consumer Devices and Services
Publication status	Published - 1 Jan 2018

Authors (Note for portal: view the doc link for the full list of authors)

Authors
Lemley, J,Bazrafkan, S,Corcoran, P,Mohanty, SP,Corcoran, P,Li, H,Sengupta, A,Lee, JH

Cite this

@inproceedings{597de596f77941bb920d0ced6f1a49ca,

title = "2018 IEEE INTERNATIONAL CONFERENCE ON CONSUMER ELECTRONICS (ICCE)",

abstract = "Transferring the success of deep learning models to consumer electronic devices requires the construction of deep learning models that are small enough to fit on resource constrained hardware. Since embedded and mobile devices lack the resources in terms of power consumption requirements, processing speed, and available memory of the latest GPU technology, it is desirable to create neural networks that are significantly smaller without sacrificing accuracy. A new technique for data augmentation called Smart Augmentation has recently been introduced that has been experimentally shown to be effective at this task. In this paper, we show how Smart Augmentation can be used to train models that are significantly smaller than their equivalently performant counterparts, and thus more viable for deployment on consumer devices.",

author = "Peter Corcoran",

year = "2018",

month = jan,

day = "1",

language = "English (Ireland)",

booktitle = "Learning data Augmentation for Consumer Devices and Services",

}

TY - GEN

T1 - 2018 IEEE INTERNATIONAL CONFERENCE ON CONSUMER ELECTRONICS (ICCE)

AU - Corcoran, Peter

PY - 2018/1/1

Y1 - 2018/1/1

N2 - Transferring the success of deep learning models to consumer electronic devices requires the construction of deep learning models that are small enough to fit on resource constrained hardware. Since embedded and mobile devices lack the resources in terms of power consumption requirements, processing speed, and available memory of the latest GPU technology, it is desirable to create neural networks that are significantly smaller without sacrificing accuracy. A new technique for data augmentation called Smart Augmentation has recently been introduced that has been experimentally shown to be effective at this task. In this paper, we show how Smart Augmentation can be used to train models that are significantly smaller than their equivalently performant counterparts, and thus more viable for deployment on consumer devices.

AB - Transferring the success of deep learning models to consumer electronic devices requires the construction of deep learning models that are small enough to fit on resource constrained hardware. Since embedded and mobile devices lack the resources in terms of power consumption requirements, processing speed, and available memory of the latest GPU technology, it is desirable to create neural networks that are significantly smaller without sacrificing accuracy. A new technique for data augmentation called Smart Augmentation has recently been introduced that has been experimentally shown to be effective at this task. In this paper, we show how Smart Augmentation can be used to train models that are significantly smaller than their equivalently performant counterparts, and thus more viable for deployment on consumer devices.

M3 - Conference Publication

BT - Learning data Augmentation for Consumer Devices and Services

ER -

2018 IEEE INTERNATIONAL CONFERENCE ON CONSUMER ELECTRONICS (ICCE)

Abstract

Authors (Note for portal: view the doc link for the full list of authors)

Fingerprint

Cite this