Tailor : Altering Skip Connections for Resource-Efficient Inference
ACM Transactions on Reconfigurable Technology and Systems · 31.03.2024 · 17 (1) · 1-23
DOI: 10.1145/3624990 ↗Kurz gefasst
Fachartikel in der Zeitschrift ACM Transactions on Reconfigurable Technology and Systems, online erschienen am 27.01.2024. Verfasst von Olivia Weng und 10 weiteren Autor:innen. Die Arbeit wurde bisher 15-mal zitiert (Crossref) und nennt 52 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Science Foundation Graduate Research Fellowship Program, National Science Foundation, U.S. Department of Energy.
Abstract (Originalsprache)
Deep neural networks use skip connections to improve training convergence. However, these skip connections are costly in hardware, requiring extra buffers and increasing on- and off-chip memory utilization and bandwidth requirements. In this article, we show that skip connections can be optimized for hardware when tackled with a hardware-software codesign approach. We argue that while a network’s skip connections are needed for the network to learn, they can later be removed or shortened to provide a more hardware-efficient implementation with minimal to no accuracy loss. We introduce Tailor , a codesign tool whose hardware-aware training algorithm gradually removes or shortens a fully trained network’s skip connections to lower the hardware cost. Tailor improves resource utilization by up to 34% for block random access memories (BRAMs), 13% for flip-flops (FFs), and 16% for look-up tables (LUTs) for on-chip, dataflow-style architectures. Tailor increases performance by 30% and reduces memory bandwidth by 45% for a two-dimensional processing element array architecture.
Förderung
- National Science Foundation Graduate Research Fellowship Program
- National Science Foundation
- U.S. Department of Energy
- Office of Science, Office of Advanced Scientific Computing Research
- DOE Office of Science, Office of High Energy Physics Early Career Research Program
Zitieren
APA
Weng, O., Marcano, G., Loncar, V., Khodamoradi, A., G, A., Sheybani, N., Meza, A., Koushanfar, F., Denolf, K., Duarte, J. M., & Kastner, R. (2024). Tailor : Altering Skip Connections for Resource-Efficient Inference. ACM Transactions on Reconfigurable Technology and Systems, 17(1), 1-23. https://doi.org/10.1145/3624990
Vancouver
Weng O, Marcano G, Loncar V, Khodamoradi A, G A, Sheybani N, et al. Tailor : Altering Skip Connections for Resource-Efficient Inference. ACM Transactions on Reconfigurable Technology and Systems. 2024;17(1):1-23. doi:10.1145/3624990
BibTeX
@article{Weng2024,
author = {Weng, Olivia and Marcano, Gabriel and Loncar, Vladimir and Khodamoradi, Alireza and G, Abarajithan and Sheybani, Nojan and Meza, Andres and Koushanfar, Farinaz and Denolf, Kristof and Duarte, Javier Mauricio and Kastner, Ryan},
title = {Tailor : Altering Skip Connections for Resource-Efficient Inference},
journal = {ACM Transactions on Reconfigurable Technology and Systems},
year = {2024},
volume = {17},
number = {1},
pages = {1--23},
doi = {10.1145/3624990}
}
EUVIMED