Linear computation coding

Müller R, Gäde B, Bereyhi A (2021)


Publication Type: Conference contribution

Publication year: 2021

Journal

Publisher: Institute of Electrical and Electronics Engineers Inc.

Book Volume: 2021-June

Pages Range: 5065-5069

Conference Proceedings Title: ICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings

Event location: Virtual, Toronto, ON, CAN

DOI: 10.1109/ICASSP39728.2021.9414317

Abstract

We introduce the new concept of computation coding. For linear functions, we present an algorithm to reduce the computational cost of multiplying an arbitrary given matrix with an unknown vector. It decomposes the given matrix into the product of codebook and wiring matrices whose entries are either zero or signed integer powers of two. For a typical implementation of deep neural networks, the proposed algorithm reduces the number of required addition units several times. To achieve the accuracy of 16-bit signed integer arithmetic for 4k-vectors, no multipliers and only 1.5 adders per matrix entry are needed.

Authors with CRIS profile

How to cite

APA:

Müller, R., Gäde, B., & Bereyhi, A. (2021). Linear computation coding. In ICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings (pp. 5065-5069). Virtual, Toronto, ON, CAN: Institute of Electrical and Electronics Engineers Inc..

MLA:

Müller, Ralf, Bernhard Gäde, and Ali Bereyhi. "Linear computation coding." Proceedings of the 2021 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2021, Virtual, Toronto, ON, CAN Institute of Electrical and Electronics Engineers Inc., 2021. 5065-5069.

BibTeX: Download