Hardware requirements for neural-net optical character recognition

L. D. Jackel, B. Boser, J. S. Denker, H. P. Graf, Yann LeCun, I. Guyon, D. Henderson, R. E. Howard, W. Hubbard, S. A. Solla

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

Abstract

Hardware architectures for character recognition are discussed, and choices for possible circuits are outlined. An advanced (and working) reconfigurable neural-net chip that mixes analog and digital processing is described. It is found that different approaches to image recognition often lead to neural-net architectures that have limited connectivity and repeated use of the same set of weights. This architecture is ideal for time-multiplexing (a combined parallel-series processing) on hardware systems that would be too small to evaluate the entire network in parallel. To make this process efficient, a chip needs to have shift registers to format the input data and additional registers to store intermediate results. Within this framework, it is possible to design chips that have broad utility, large connection capacity, and high speed. This was demonstrated by a new chip with 32,000 reconfigurable connections.

Original languageEnglish (US)
Title of host publicationIJCNN. International Joint Conference on Neural Networks
PublisherPubl by IEEE
Pages855-861
Number of pages7
StatePublished - 1990
Event1990 International Joint Conference on Neural Networks - IJCNN 90 Part 3 (of 3) - San Diego, CA, USA
Duration: Jun 17 1990Jun 21 1990

Other

Other1990 International Joint Conference on Neural Networks - IJCNN 90 Part 3 (of 3)
CitySan Diego, CA, USA
Period6/17/906/21/90

Fingerprint

Optical character recognition
Neural networks
Hardware
Image recognition
Shift registers
Character recognition
Digital signal processing
Multiplexing
Networks (circuits)
Processing

ASJC Scopus subject areas

  • Engineering(all)

Cite this

Jackel, L. D., Boser, B., Denker, J. S., Graf, H. P., LeCun, Y., Guyon, I., ... Solla, S. A. (1990). Hardware requirements for neural-net optical character recognition. In IJCNN. International Joint Conference on Neural Networks (pp. 855-861). Publ by IEEE.

Hardware requirements for neural-net optical character recognition. / Jackel, L. D.; Boser, B.; Denker, J. S.; Graf, H. P.; LeCun, Yann; Guyon, I.; Henderson, D.; Howard, R. E.; Hubbard, W.; Solla, S. A.

IJCNN. International Joint Conference on Neural Networks. Publ by IEEE, 1990. p. 855-861.

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

Jackel, LD, Boser, B, Denker, JS, Graf, HP, LeCun, Y, Guyon, I, Henderson, D, Howard, RE, Hubbard, W & Solla, SA 1990, Hardware requirements for neural-net optical character recognition. in IJCNN. International Joint Conference on Neural Networks. Publ by IEEE, pp. 855-861, 1990 International Joint Conference on Neural Networks - IJCNN 90 Part 3 (of 3), San Diego, CA, USA, 6/17/90.
Jackel LD, Boser B, Denker JS, Graf HP, LeCun Y, Guyon I et al. Hardware requirements for neural-net optical character recognition. In IJCNN. International Joint Conference on Neural Networks. Publ by IEEE. 1990. p. 855-861
Jackel, L. D. ; Boser, B. ; Denker, J. S. ; Graf, H. P. ; LeCun, Yann ; Guyon, I. ; Henderson, D. ; Howard, R. E. ; Hubbard, W. ; Solla, S. A. / Hardware requirements for neural-net optical character recognition. IJCNN. International Joint Conference on Neural Networks. Publ by IEEE, 1990. pp. 855-861
@inproceedings{8116faddb08541d3ac8caa8a7a82c922,
title = "Hardware requirements for neural-net optical character recognition",
abstract = "Hardware architectures for character recognition are discussed, and choices for possible circuits are outlined. An advanced (and working) reconfigurable neural-net chip that mixes analog and digital processing is described. It is found that different approaches to image recognition often lead to neural-net architectures that have limited connectivity and repeated use of the same set of weights. This architecture is ideal for time-multiplexing (a combined parallel-series processing) on hardware systems that would be too small to evaluate the entire network in parallel. To make this process efficient, a chip needs to have shift registers to format the input data and additional registers to store intermediate results. Within this framework, it is possible to design chips that have broad utility, large connection capacity, and high speed. This was demonstrated by a new chip with 32,000 reconfigurable connections.",
author = "Jackel, {L. D.} and B. Boser and Denker, {J. S.} and Graf, {H. P.} and Yann LeCun and I. Guyon and D. Henderson and Howard, {R. E.} and W. Hubbard and Solla, {S. A.}",
year = "1990",
language = "English (US)",
pages = "855--861",
booktitle = "IJCNN. International Joint Conference on Neural Networks",
publisher = "Publ by IEEE",

}

TY - GEN

T1 - Hardware requirements for neural-net optical character recognition

AU - Jackel, L. D.

AU - Boser, B.

AU - Denker, J. S.

AU - Graf, H. P.

AU - LeCun, Yann

AU - Guyon, I.

AU - Henderson, D.

AU - Howard, R. E.

AU - Hubbard, W.

AU - Solla, S. A.

PY - 1990

Y1 - 1990

N2 - Hardware architectures for character recognition are discussed, and choices for possible circuits are outlined. An advanced (and working) reconfigurable neural-net chip that mixes analog and digital processing is described. It is found that different approaches to image recognition often lead to neural-net architectures that have limited connectivity and repeated use of the same set of weights. This architecture is ideal for time-multiplexing (a combined parallel-series processing) on hardware systems that would be too small to evaluate the entire network in parallel. To make this process efficient, a chip needs to have shift registers to format the input data and additional registers to store intermediate results. Within this framework, it is possible to design chips that have broad utility, large connection capacity, and high speed. This was demonstrated by a new chip with 32,000 reconfigurable connections.

AB - Hardware architectures for character recognition are discussed, and choices for possible circuits are outlined. An advanced (and working) reconfigurable neural-net chip that mixes analog and digital processing is described. It is found that different approaches to image recognition often lead to neural-net architectures that have limited connectivity and repeated use of the same set of weights. This architecture is ideal for time-multiplexing (a combined parallel-series processing) on hardware systems that would be too small to evaluate the entire network in parallel. To make this process efficient, a chip needs to have shift registers to format the input data and additional registers to store intermediate results. Within this framework, it is possible to design chips that have broad utility, large connection capacity, and high speed. This was demonstrated by a new chip with 32,000 reconfigurable connections.

UR - http://www.scopus.com/inward/record.url?scp=0025547059&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=0025547059&partnerID=8YFLogxK

M3 - Conference contribution

AN - SCOPUS:0025547059

SP - 855

EP - 861

BT - IJCNN. International Joint Conference on Neural Networks

PB - Publ by IEEE

ER -