Co-Training of Feature Extraction and Classification using Partitioned Convolutional Neural Networks

Wei Yu Tsai, Jinhang Choi, Tulika Parija, Priyanka Gomatam, Chitaranjan Das, John Morgan Sampson, Vijaykrishnan Narayanan

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

1 Citation (Scopus)

Abstract

There are an increasing number of neuromorphic hardware platforms designed to efficiently support neural network inference tasks. However, many applications contain structured processing in addition to classification. Being able to map both neural network classification and structured computation onto the same platform is appealing from a system design perspective. In this paper, we perform a case study on mapping the feature extraction stage of pedestrian detection using Histogram of Oriented Gradients (HoG) onto a neuromophic platform. We consider three implementations: one that approximates HoG using neuromorphic intrinsics, one that emulates HoG outputs using a trained network, and one that allows feature extraction to be absorbed into classification. The proposed feature extraction methods are implemented and evaluated on neuromorphic hardware (IBM Neurosynaptic System). Our study shows that both a designed approximation and a "parroted" emulation can achieve similar accuracy, and that the latter appears to better capitalize on limited training and resource budgets, compared to the absorbed approach, while also being more power efficient than the programmed approach by a factor of 6.5x-208x.

Original languageEnglish (US)
Title of host publicationProceedings of the 54th Annual Design Automation Conference 2017, DAC 2017
PublisherInstitute of Electrical and Electronics Engineers Inc.
ISBN (Electronic)9781450349277
DOIs
StatePublished - Jun 18 2017
Event54th Annual Design Automation Conference, DAC 2017 - Austin, United States
Duration: Jun 18 2017Jun 22 2017

Publication series

NameProceedings - Design Automation Conference
VolumePart 128280
ISSN (Print)0738-100X

Other

Other54th Annual Design Automation Conference, DAC 2017
CountryUnited States
CityAustin
Period6/18/176/22/17

Fingerprint

Co-training
Histogram
Feature Extraction
Feature extraction
Neural Networks
Gradient
Neural networks
Hardware
Pedestrian Detection
Emulation
System Design
Systems analysis
Resources
Output
Approximation
Processing

All Science Journal Classification (ASJC) codes

  • Computer Science Applications
  • Control and Systems Engineering
  • Electrical and Electronic Engineering
  • Modeling and Simulation

Cite this

Tsai, W. Y., Choi, J., Parija, T., Gomatam, P., Das, C., Sampson, J. M., & Narayanan, V. (2017). Co-Training of Feature Extraction and Classification using Partitioned Convolutional Neural Networks. In Proceedings of the 54th Annual Design Automation Conference 2017, DAC 2017 [58] (Proceedings - Design Automation Conference; Vol. Part 128280). Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1145/3061639.3062218
Tsai, Wei Yu ; Choi, Jinhang ; Parija, Tulika ; Gomatam, Priyanka ; Das, Chitaranjan ; Sampson, John Morgan ; Narayanan, Vijaykrishnan. / Co-Training of Feature Extraction and Classification using Partitioned Convolutional Neural Networks. Proceedings of the 54th Annual Design Automation Conference 2017, DAC 2017. Institute of Electrical and Electronics Engineers Inc., 2017. (Proceedings - Design Automation Conference).
@inproceedings{5253c2573f204e2bbc429bc1e40ba9b7,
title = "Co-Training of Feature Extraction and Classification using Partitioned Convolutional Neural Networks",
abstract = "There are an increasing number of neuromorphic hardware platforms designed to efficiently support neural network inference tasks. However, many applications contain structured processing in addition to classification. Being able to map both neural network classification and structured computation onto the same platform is appealing from a system design perspective. In this paper, we perform a case study on mapping the feature extraction stage of pedestrian detection using Histogram of Oriented Gradients (HoG) onto a neuromophic platform. We consider three implementations: one that approximates HoG using neuromorphic intrinsics, one that emulates HoG outputs using a trained network, and one that allows feature extraction to be absorbed into classification. The proposed feature extraction methods are implemented and evaluated on neuromorphic hardware (IBM Neurosynaptic System). Our study shows that both a designed approximation and a {"}parroted{"} emulation can achieve similar accuracy, and that the latter appears to better capitalize on limited training and resource budgets, compared to the absorbed approach, while also being more power efficient than the programmed approach by a factor of 6.5x-208x.",
author = "Tsai, {Wei Yu} and Jinhang Choi and Tulika Parija and Priyanka Gomatam and Chitaranjan Das and Sampson, {John Morgan} and Vijaykrishnan Narayanan",
year = "2017",
month = "6",
day = "18",
doi = "10.1145/3061639.3062218",
language = "English (US)",
series = "Proceedings - Design Automation Conference",
publisher = "Institute of Electrical and Electronics Engineers Inc.",
booktitle = "Proceedings of the 54th Annual Design Automation Conference 2017, DAC 2017",
address = "United States",

}

Tsai, WY, Choi, J, Parija, T, Gomatam, P, Das, C, Sampson, JM & Narayanan, V 2017, Co-Training of Feature Extraction and Classification using Partitioned Convolutional Neural Networks. in Proceedings of the 54th Annual Design Automation Conference 2017, DAC 2017., 58, Proceedings - Design Automation Conference, vol. Part 128280, Institute of Electrical and Electronics Engineers Inc., 54th Annual Design Automation Conference, DAC 2017, Austin, United States, 6/18/17. https://doi.org/10.1145/3061639.3062218

Co-Training of Feature Extraction and Classification using Partitioned Convolutional Neural Networks. / Tsai, Wei Yu; Choi, Jinhang; Parija, Tulika; Gomatam, Priyanka; Das, Chitaranjan; Sampson, John Morgan; Narayanan, Vijaykrishnan.

Proceedings of the 54th Annual Design Automation Conference 2017, DAC 2017. Institute of Electrical and Electronics Engineers Inc., 2017. 58 (Proceedings - Design Automation Conference; Vol. Part 128280).

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

TY - GEN

T1 - Co-Training of Feature Extraction and Classification using Partitioned Convolutional Neural Networks

AU - Tsai, Wei Yu

AU - Choi, Jinhang

AU - Parija, Tulika

AU - Gomatam, Priyanka

AU - Das, Chitaranjan

AU - Sampson, John Morgan

AU - Narayanan, Vijaykrishnan

PY - 2017/6/18

Y1 - 2017/6/18

N2 - There are an increasing number of neuromorphic hardware platforms designed to efficiently support neural network inference tasks. However, many applications contain structured processing in addition to classification. Being able to map both neural network classification and structured computation onto the same platform is appealing from a system design perspective. In this paper, we perform a case study on mapping the feature extraction stage of pedestrian detection using Histogram of Oriented Gradients (HoG) onto a neuromophic platform. We consider three implementations: one that approximates HoG using neuromorphic intrinsics, one that emulates HoG outputs using a trained network, and one that allows feature extraction to be absorbed into classification. The proposed feature extraction methods are implemented and evaluated on neuromorphic hardware (IBM Neurosynaptic System). Our study shows that both a designed approximation and a "parroted" emulation can achieve similar accuracy, and that the latter appears to better capitalize on limited training and resource budgets, compared to the absorbed approach, while also being more power efficient than the programmed approach by a factor of 6.5x-208x.

AB - There are an increasing number of neuromorphic hardware platforms designed to efficiently support neural network inference tasks. However, many applications contain structured processing in addition to classification. Being able to map both neural network classification and structured computation onto the same platform is appealing from a system design perspective. In this paper, we perform a case study on mapping the feature extraction stage of pedestrian detection using Histogram of Oriented Gradients (HoG) onto a neuromophic platform. We consider three implementations: one that approximates HoG using neuromorphic intrinsics, one that emulates HoG outputs using a trained network, and one that allows feature extraction to be absorbed into classification. The proposed feature extraction methods are implemented and evaluated on neuromorphic hardware (IBM Neurosynaptic System). Our study shows that both a designed approximation and a "parroted" emulation can achieve similar accuracy, and that the latter appears to better capitalize on limited training and resource budgets, compared to the absorbed approach, while also being more power efficient than the programmed approach by a factor of 6.5x-208x.

UR - http://www.scopus.com/inward/record.url?scp=85023595603&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85023595603&partnerID=8YFLogxK

U2 - 10.1145/3061639.3062218

DO - 10.1145/3061639.3062218

M3 - Conference contribution

AN - SCOPUS:85023595603

T3 - Proceedings - Design Automation Conference

BT - Proceedings of the 54th Annual Design Automation Conference 2017, DAC 2017

PB - Institute of Electrical and Electronics Engineers Inc.

ER -

Tsai WY, Choi J, Parija T, Gomatam P, Das C, Sampson JM et al. Co-Training of Feature Extraction and Classification using Partitioned Convolutional Neural Networks. In Proceedings of the 54th Annual Design Automation Conference 2017, DAC 2017. Institute of Electrical and Electronics Engineers Inc. 2017. 58. (Proceedings - Design Automation Conference). https://doi.org/10.1145/3061639.3062218