Design Space Exploration of Time, Energy, and Error Rate Trade-offs for CNNs using Accuracy-Programmable Instruction Set Processors

Schuster A, Heidorn C, Brand M, Keszöcze O, Teich J (2021)

Publication Type: Conference contribution, Conference Contribution

Publication year: 2021

Publisher: Springer Nature

Series: Communications in Computer and Information Science

City/Town: Switzerland

Book Volume: 1524

Pages Range: 375-389

Conference Proceedings Title: Joint European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases (ECML PKDD 2021)

Event location: Virtual Event

ISBN: 978-3-030-93736-2

DOI: 10.1007/978-3-030-93736-2_29

Abstract

We proclaim the use of application-specific instruction set processors with programmable accuracy called Anytime Instruction Processors (AIPs) for Convolutional Neural Network (CNN) inference. For a floating-point operation, the number of correctly computed mantissa result bits can be freely adjusted, allowing for a fine-grained trade-off analysis between accuracy, execution time and energy. We propose a Design Space Exploration (DSE) technique in which the accuracy of CNN computations is determined layer-by-layer. As one result, we show that reductions of up to 62% in energy consumption are achievable for a representative ResNet-18 benchmark in comparison to a solution where each layer is computed at full accuracy according to the IEEE 754 single precision floating-point format.

Authors with CRIS profile

Armin Schuster Lehrstuhl für Informatik 12 (Hardware-Software-Co-Design) Christian Heidorn Lehrstuhl für Informatik 12 (Hardware-Software-Co-Design) Marcel Brand Lehrstuhl für Informatik 12 (Hardware-Software-Co-Design) Oliver Keszöcze Juniorprofessur für Informatik Jürgen Teich Lehrstuhl für Informatik 12 (Hardware-Software-Co-Design)

Related research project(s)

ACoF -- Approximate Computing on FPGAs (ACoF) June 1, 2021 - June 1, 2024 DFG SFB/Transregio 89 "Invasive Computing" (TRR 89) July 1, 2010 - Dec. 31, 2022 AI Laboratory for System-level Design of ML-based Signal Processing Applications (KISS) Oct. 1, 2019 - Jan. 31, 2022

How to cite

APA:

Schuster, A., Heidorn, C., Brand, M., Keszöcze, O., & Teich, J. (2021). Design Space Exploration of Time, Energy, and Error Rate Trade-offs for CNNs using Accuracy-Programmable Instruction Set Processors. In Springer, Cham (Eds.), Joint European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases (ECML PKDD 2021) (pp. 375-389). Virtual Event: Switzerland: Springer Nature.

MLA:

Schuster, Armin, et al. "Design Space Exploration of Time, Energy, and Error Rate Trade-offs for CNNs using Accuracy-Programmable Instruction Set Processors." Proceedings of the 2nd International Workshop on IoT, Edge, and Mobile for Embedded Machine Learning (ITEM), Virtual Event Ed. Springer, Cham, Switzerland: Springer Nature, 2021. 375-389.

BibTeX: Download