Conditional Deep Learning for energy-efficient and enhanced pattern recognition

Priyadarshini Panda, Abhronil Sengupta, Kaushik Roy

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

21 Citations (Scopus)

Abstract

Deep learning neural networks have emerged as one of the most powerful classification tools for vision related applications. However, the computational and energy requirements associated with such deep nets can be quite high, and hence their energy-efficient implementation is of great interest. Although traditionally the entire network is utilized for the recognition of all inputs, we observe that the classification difficulty varies widely across inputs in real-world datasets; only a small fraction of inputs require the full computational effort of a network, while a large majority can be classified correctly with very low effort. In this paper, we propose Conditional Deep Learning (CDL) where the convolutional layer features are used to identify the variability in the difficulty of input instances and conditionally activate the deeper layers of the network. We achieve this by cascading a linear network of output neurons for each convolutional layer and monitoring the output of the linear network to decide whether classification can be terminated at the current stage or not. The proposed methodology thus enables the network to dynamically adjust the computational effort depending upon the difficulty of the input data while maintaining competitive classification accuracy. We evaluate our approach on the MNIST dataset. Our experiments demonstrate that our proposed CDL yields 1.91× reduction in average number of operations per input, which translates to 1.84× improvement in energy. In addition, our results show an improvement in classification accuracy from 97.5% to 98.9% as compared to the original network.

Original languageEnglish (US)
Title of host publicationProceedings of the 2016 Design, Automation and Test in Europe Conference and Exhibition, DATE 2016
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages475-480
Number of pages6
ISBN (Electronic)9783981537062
StatePublished - Apr 25 2016
Event19th Design, Automation and Test in Europe Conference and Exhibition, DATE 2016 - Dresden, Germany
Duration: Mar 14 2016Mar 18 2016

Publication series

NameProceedings of the 2016 Design, Automation and Test in Europe Conference and Exhibition, DATE 2016

Conference

Conference19th Design, Automation and Test in Europe Conference and Exhibition, DATE 2016
CountryGermany
CityDresden
Period3/14/163/18/16

Fingerprint

Pattern recognition
Linear networks
Neurons
Deep learning
Neural networks
Monitoring
Experiments

All Science Journal Classification (ASJC) codes

  • Hardware and Architecture
  • Safety, Risk, Reliability and Quality

Cite this

Panda, P., Sengupta, A., & Roy, K. (2016). Conditional Deep Learning for energy-efficient and enhanced pattern recognition. In Proceedings of the 2016 Design, Automation and Test in Europe Conference and Exhibition, DATE 2016 (pp. 475-480). [7459357] (Proceedings of the 2016 Design, Automation and Test in Europe Conference and Exhibition, DATE 2016). Institute of Electrical and Electronics Engineers Inc..
Panda, Priyadarshini ; Sengupta, Abhronil ; Roy, Kaushik. / Conditional Deep Learning for energy-efficient and enhanced pattern recognition. Proceedings of the 2016 Design, Automation and Test in Europe Conference and Exhibition, DATE 2016. Institute of Electrical and Electronics Engineers Inc., 2016. pp. 475-480 (Proceedings of the 2016 Design, Automation and Test in Europe Conference and Exhibition, DATE 2016).
@inproceedings{8937379d06dd48f89602bbeedeb69c17,
title = "Conditional Deep Learning for energy-efficient and enhanced pattern recognition",
abstract = "Deep learning neural networks have emerged as one of the most powerful classification tools for vision related applications. However, the computational and energy requirements associated with such deep nets can be quite high, and hence their energy-efficient implementation is of great interest. Although traditionally the entire network is utilized for the recognition of all inputs, we observe that the classification difficulty varies widely across inputs in real-world datasets; only a small fraction of inputs require the full computational effort of a network, while a large majority can be classified correctly with very low effort. In this paper, we propose Conditional Deep Learning (CDL) where the convolutional layer features are used to identify the variability in the difficulty of input instances and conditionally activate the deeper layers of the network. We achieve this by cascading a linear network of output neurons for each convolutional layer and monitoring the output of the linear network to decide whether classification can be terminated at the current stage or not. The proposed methodology thus enables the network to dynamically adjust the computational effort depending upon the difficulty of the input data while maintaining competitive classification accuracy. We evaluate our approach on the MNIST dataset. Our experiments demonstrate that our proposed CDL yields 1.91× reduction in average number of operations per input, which translates to 1.84× improvement in energy. In addition, our results show an improvement in classification accuracy from 97.5{\%} to 98.9{\%} as compared to the original network.",
author = "Priyadarshini Panda and Abhronil Sengupta and Kaushik Roy",
year = "2016",
month = "4",
day = "25",
language = "English (US)",
series = "Proceedings of the 2016 Design, Automation and Test in Europe Conference and Exhibition, DATE 2016",
publisher = "Institute of Electrical and Electronics Engineers Inc.",
pages = "475--480",
booktitle = "Proceedings of the 2016 Design, Automation and Test in Europe Conference and Exhibition, DATE 2016",
address = "United States",

}

Panda, P, Sengupta, A & Roy, K 2016, Conditional Deep Learning for energy-efficient and enhanced pattern recognition. in Proceedings of the 2016 Design, Automation and Test in Europe Conference and Exhibition, DATE 2016., 7459357, Proceedings of the 2016 Design, Automation and Test in Europe Conference and Exhibition, DATE 2016, Institute of Electrical and Electronics Engineers Inc., pp. 475-480, 19th Design, Automation and Test in Europe Conference and Exhibition, DATE 2016, Dresden, Germany, 3/14/16.

Conditional Deep Learning for energy-efficient and enhanced pattern recognition. / Panda, Priyadarshini; Sengupta, Abhronil; Roy, Kaushik.

Proceedings of the 2016 Design, Automation and Test in Europe Conference and Exhibition, DATE 2016. Institute of Electrical and Electronics Engineers Inc., 2016. p. 475-480 7459357 (Proceedings of the 2016 Design, Automation and Test in Europe Conference and Exhibition, DATE 2016).

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

TY - GEN

T1 - Conditional Deep Learning for energy-efficient and enhanced pattern recognition

AU - Panda, Priyadarshini

AU - Sengupta, Abhronil

AU - Roy, Kaushik

PY - 2016/4/25

Y1 - 2016/4/25

N2 - Deep learning neural networks have emerged as one of the most powerful classification tools for vision related applications. However, the computational and energy requirements associated with such deep nets can be quite high, and hence their energy-efficient implementation is of great interest. Although traditionally the entire network is utilized for the recognition of all inputs, we observe that the classification difficulty varies widely across inputs in real-world datasets; only a small fraction of inputs require the full computational effort of a network, while a large majority can be classified correctly with very low effort. In this paper, we propose Conditional Deep Learning (CDL) where the convolutional layer features are used to identify the variability in the difficulty of input instances and conditionally activate the deeper layers of the network. We achieve this by cascading a linear network of output neurons for each convolutional layer and monitoring the output of the linear network to decide whether classification can be terminated at the current stage or not. The proposed methodology thus enables the network to dynamically adjust the computational effort depending upon the difficulty of the input data while maintaining competitive classification accuracy. We evaluate our approach on the MNIST dataset. Our experiments demonstrate that our proposed CDL yields 1.91× reduction in average number of operations per input, which translates to 1.84× improvement in energy. In addition, our results show an improvement in classification accuracy from 97.5% to 98.9% as compared to the original network.

AB - Deep learning neural networks have emerged as one of the most powerful classification tools for vision related applications. However, the computational and energy requirements associated with such deep nets can be quite high, and hence their energy-efficient implementation is of great interest. Although traditionally the entire network is utilized for the recognition of all inputs, we observe that the classification difficulty varies widely across inputs in real-world datasets; only a small fraction of inputs require the full computational effort of a network, while a large majority can be classified correctly with very low effort. In this paper, we propose Conditional Deep Learning (CDL) where the convolutional layer features are used to identify the variability in the difficulty of input instances and conditionally activate the deeper layers of the network. We achieve this by cascading a linear network of output neurons for each convolutional layer and monitoring the output of the linear network to decide whether classification can be terminated at the current stage or not. The proposed methodology thus enables the network to dynamically adjust the computational effort depending upon the difficulty of the input data while maintaining competitive classification accuracy. We evaluate our approach on the MNIST dataset. Our experiments demonstrate that our proposed CDL yields 1.91× reduction in average number of operations per input, which translates to 1.84× improvement in energy. In addition, our results show an improvement in classification accuracy from 97.5% to 98.9% as compared to the original network.

UR - http://www.scopus.com/inward/record.url?scp=84973643647&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84973643647&partnerID=8YFLogxK

M3 - Conference contribution

AN - SCOPUS:84973643647

T3 - Proceedings of the 2016 Design, Automation and Test in Europe Conference and Exhibition, DATE 2016

SP - 475

EP - 480

BT - Proceedings of the 2016 Design, Automation and Test in Europe Conference and Exhibition, DATE 2016

PB - Institute of Electrical and Electronics Engineers Inc.

ER -

Panda P, Sengupta A, Roy K. Conditional Deep Learning for energy-efficient and enhanced pattern recognition. In Proceedings of the 2016 Design, Automation and Test in Europe Conference and Exhibition, DATE 2016. Institute of Electrical and Electronics Engineers Inc. 2016. p. 475-480. 7459357. (Proceedings of the 2016 Design, Automation and Test in Europe Conference and Exhibition, DATE 2016).