Approximate FPGA-based LSTMs under computation time constraints
File(s)
Accepted version
Author(s)
Rizakis, Michalis
Venieris, Stylianos I
Kouris, Alexandros
Bouganis, Christos-Savvas
Type
Conference Paper
Abstract
Recurrent Neural Networks, with the prominence of Long
Short-Term Memory (LSTM) networks, have demonstrated state-of-the-
art accuracy in several emerging Artificial Intelligence tasks. Neverthe-
less, the highest performing LSTM models are becoming increasingly
demanding in terms of computational and memory load. At the same
time, emerging latency-sensitive applications including mobile robots and
autonomous vehicles often operate under stringent computation time
constraints. In this paper, we address the challenge of deploying com-
putationally demanding LSTMs at a constrained time budget by intro-
ducing an approximate computing scheme that combines iterative low-
rank compression and pruning, along with a novel FPGA-based LSTM
architecture. Combined in an end-to-end framework, the approximation
method parameters are optimised and the architecture is configured
to address the problem of high-performance LSTM execution in time-
constrained applications. Quantitative evaluation on a real-life image
captioning application indicates that the proposed system required up to
6.5
×
less time to achieve the same application-level accuracy compared
to a baseline method, while achieving an average of 25
×
higher accuracy
under the same computation time constraints.
Short-Term Memory (LSTM) networks, have demonstrated state-of-the-
art accuracy in several emerging Artificial Intelligence tasks. Neverthe-
less, the highest performing LSTM models are becoming increasingly
demanding in terms of computational and memory load. At the same
time, emerging latency-sensitive applications including mobile robots and
autonomous vehicles often operate under stringent computation time
constraints. In this paper, we address the challenge of deploying com-
putationally demanding LSTMs at a constrained time budget by intro-
ducing an approximate computing scheme that combines iterative low-
rank compression and pruning, along with a novel FPGA-based LSTM
architecture. Combined in an end-to-end framework, the approximation
method parameters are optimised and the architecture is configured
to address the problem of high-performance LSTM execution in time-
constrained applications. Quantitative evaluation on a real-life image
captioning application indicates that the proposed system required up to
6.5
×
less time to achieve the same application-level accuracy compared
to a baseline method, while achieving an average of 25
×
higher accuracy
under the same computation time constraints.
Editor(s)
Voros, Nikolaos S
Hübner, Michael
Keramidas, Georgios
Goehringer, Diana
Antonopoulos, Christos P
Diniz, Pedro C
Date Issued
2018-05-02
Date Acceptance
2018-05-02
Citation
Applied Reconfigurable Computing. Architectures, Tools, and Applications, 2018, 10824, pp.3-15
ISBN
9783319788890
ISSN
0302-9743
Publisher
Springer
Start Page
3
End Page
15
Journal / Book Title
Applied Reconfigurable Computing. Architectures, Tools, and Applications
Volume
10824
Copyright Statement
© 2018 Springer International Publishing AG, part of Springer Nature. The final publication is available at Springer via https://dx.doi.org/10.1007/978-3-319-78890-6
Identifier
http://arxiv.org/abs/1801.02190v1
Source
ARC 2018: 14th International Symposium on Applied Reconfigurable Computing
Subjects
cs.CV
cs.CV
cs.AR
cs.LG
Publication Status
Published
Start Date
2018-05-02
Finish Date
2018-05-04
Coverage Spatial
Santorini, Greece