Publication detail

Inference Energy Analysis in Context of Hardware-Aware NAS

PIŇOS, M. KLHŮFEK, J. MRÁZEK, V. SEKANINA, L.

Original Title

Inference Energy Analysis in Context of Hardware-Aware NAS

Type

conference paper

Language

English

Original Abstract

Hardware-aware neural architecture search (HW-aware NAS) methods are crucial for designing and optimizing deep neural networks (DNNs) for efficient deployment on hardware accelerators. In this work, we analyze two HW-aware NAS methods, EvoApproxNAS and ApproxDARTS, and investigate the impact of precise hardware parameters (such as energy) measurement using Timeloop on their performance. In particular, we compare this precise measurement approach with the original approach employed by EvoApproxNAS and ApproxDARTS, which relied on a simple analytical energy estimation based on the number of multiplications performed during the inference phase of the convolutional neural network (CNN). Our analysis demonstrates how the improved energy measurements enhance the search process of HW-aware NAS methods, resulting in more energy-efficient architectures. Furthermore, we highlight the importance of precise hardware parameters measurement, showing that accurate hardware modeling is critical for obtaining CNNs with good accuracy-energy trade-offs. Our results show, that without precise hardware parameter measurement, the HW-aware NAS can produce acceptable results but may fail to fully exploit the potential of hardware accelerator, especially if the 8xN-bit approximate multipliers are considered, ultimately limiting the efficiency of designed architectures.

Keywords

Quantization,Estimation,Energy measurement,Neural architecture search,Convolutional neural networks,Hardware acceleration

Authors

PIŇOS, M.; KLHŮFEK, J.; MRÁZEK, V.; SEKANINA, L.

Released

1. 4. 2025

Publisher

Institute of Electrical and Electronics Engineers

Location

Lyon

ISBN

979-8-3315-2801-0

Book

2025 28th International Symposium on Design and Diagnostics of Electronic Circuits and Systems

Pages from

161

Pages to

164

Pages count

6

BibTex

@inproceedings{BUT196475,
  author="Michal {Piňos} and Jan {Klhůfek} and Vojtěch {Mrázek} and Lukáš {Sekanina}",
  title="Inference Energy Analysis in Context of Hardware-Aware NAS",
  booktitle="2025 28th International Symposium on Design and Diagnostics of Electronic Circuits and Systems",
  year="2025",
  pages="161--164",
  publisher="Institute of Electrical and Electronics Engineers",
  address="Lyon",
  doi="10.1109/DDECS63720.2025.11006674",
  isbn="979-8-3315-2801-0"
}