Enabling Efficient Inference of Convolutional Neural Networks via Approximation

Georgios Zervakis; Iraklis Anagnostopoulos; Hussam Amrouch; Jörg Henkel

doi:10.1007/978-3-030-98347-5_17

Enabling Efficient Inference of Convolutional Neural Networks via Approximation

Georgios Zervakis
, Iraklis Anagnostopoulos
, Hussam Amrouch
, Jörg Henkel

Humanoid Technologies Lab (H2T)
Southern Illinois University Carbondale
Universität Stuttgart

Research output: Chapter in Book/Report/Conference proceeding › Chapter › peer-review

Abstract

With the rapid advancements in the Artificial Intelligence area, Neural Networks (NNs) became the driving force both in general purpose and embedded computing domains. Especially, resource constrained embedded systems progressively rely on multiple NNs to provide on the spot sophisticated services. Nevertheless, supporting NN-based workloads is challenging due to the enormous computational and energy requirements. Exploiting the inherent error resiliency of NNs, significant research focuses on designing approximate Convolutional NN (CNN) inference accelerators, demonstrating that, for negligible accuracy loss, they satisfy tight latency, power, and temperature constraints. This chapter provides a comprehensive discussion of different aspects of approximate CNN implementations.

Original language	English
Title of host publication	Approximate Computing
Publisher	Springer International Publishing
Pages	429-450
Number of pages	22
ISBN (Electronic)	9783030983475
ISBN (Print)	9783030983468
DOIs	https://doi.org/10.1007/978-3-030-98347-5_17
State	Published - 1 Jan 2022
Externally published	Yes

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

SDG 7 Affordable and Clean Energy

Keywords

Approximate computing
Approximate multiplier
Circuit aging
Control variate
Convolution
Convolutional neural network (CNN)
Deep neural network (DNN)
Energy efficiency
Error analysis
Error correction
Error estimation
Inference
Low-power
Neural Processing Unit (NPU)
Quantization
Reconfigurable approximation
Reliability-aware approximation
Software-hardware codesign
Temperature-aware approximation
Thermal management

Access to Document

10.1007/978-3-030-98347-5_17

Cite this

@inbook{d70c5614cc064a2f811a8d1bf35d5ce8,

title = "Enabling Efficient Inference of Convolutional Neural Networks via Approximation",

abstract = "With the rapid advancements in the Artificial Intelligence area, Neural Networks (NNs) became the driving force both in general purpose and embedded computing domains. Especially, resource constrained embedded systems progressively rely on multiple NNs to provide on the spot sophisticated services. Nevertheless, supporting NN-based workloads is challenging due to the enormous computational and energy requirements. Exploiting the inherent error resiliency of NNs, significant research focuses on designing approximate Convolutional NN (CNN) inference accelerators, demonstrating that, for negligible accuracy loss, they satisfy tight latency, power, and temperature constraints. This chapter provides a comprehensive discussion of different aspects of approximate CNN implementations.",

keywords = "Approximate computing, Approximate multiplier, Circuit aging, Control variate, Convolution, Convolutional neural network (CNN), Deep neural network (DNN), Energy efficiency, Error analysis, Error correction, Error estimation, Inference, Low-power, Neural Processing Unit (NPU), Quantization, Reconfigurable approximation, Reliability-aware approximation, Software-hardware codesign, Temperature-aware approximation, Thermal management",

author = "Georgios Zervakis and Iraklis Anagnostopoulos and Hussam Amrouch and J{\"o}rg Henkel",

note = "Publisher Copyright: {\textcopyright} The Author(s), under exclusive license to Springer Nature Switzerland AG 2022.",

year = "2022",

month = jan,

day = "1",

doi = "10.1007/978-3-030-98347-5\_17",

language = "English",

isbn = "9783030983468",

pages = "429--450",

booktitle = "Approximate Computing",

publisher = "Springer International Publishing",

}

TY - CHAP

T1 - Enabling Efficient Inference of Convolutional Neural Networks via Approximation

AU - Zervakis, Georgios

AU - Anagnostopoulos, Iraklis

AU - Amrouch, Hussam

AU - Henkel, Jörg

N1 - Publisher Copyright: © The Author(s), under exclusive license to Springer Nature Switzerland AG 2022.

PY - 2022/1/1

Y1 - 2022/1/1

N2 - With the rapid advancements in the Artificial Intelligence area, Neural Networks (NNs) became the driving force both in general purpose and embedded computing domains. Especially, resource constrained embedded systems progressively rely on multiple NNs to provide on the spot sophisticated services. Nevertheless, supporting NN-based workloads is challenging due to the enormous computational and energy requirements. Exploiting the inherent error resiliency of NNs, significant research focuses on designing approximate Convolutional NN (CNN) inference accelerators, demonstrating that, for negligible accuracy loss, they satisfy tight latency, power, and temperature constraints. This chapter provides a comprehensive discussion of different aspects of approximate CNN implementations.

AB - With the rapid advancements in the Artificial Intelligence area, Neural Networks (NNs) became the driving force both in general purpose and embedded computing domains. Especially, resource constrained embedded systems progressively rely on multiple NNs to provide on the spot sophisticated services. Nevertheless, supporting NN-based workloads is challenging due to the enormous computational and energy requirements. Exploiting the inherent error resiliency of NNs, significant research focuses on designing approximate Convolutional NN (CNN) inference accelerators, demonstrating that, for negligible accuracy loss, they satisfy tight latency, power, and temperature constraints. This chapter provides a comprehensive discussion of different aspects of approximate CNN implementations.

KW - Approximate computing

KW - Approximate multiplier

KW - Circuit aging

KW - Control variate

KW - Convolution

KW - Convolutional neural network (CNN)

KW - Deep neural network (DNN)

KW - Energy efficiency

KW - Error analysis

KW - Error correction

KW - Error estimation

KW - Inference

KW - Low-power

KW - Neural Processing Unit (NPU)

KW - Quantization

KW - Reconfigurable approximation

KW - Reliability-aware approximation

KW - Software-hardware codesign

KW - Temperature-aware approximation

KW - Thermal management

UR - https://www.scopus.com/pages/publications/85153436249

U2 - 10.1007/978-3-030-98347-5_17

DO - 10.1007/978-3-030-98347-5_17

M3 - Chapter

AN - SCOPUS:85153436249

SN - 9783030983468

SP - 429

EP - 450

BT - Approximate Computing

PB - Springer International Publishing

ER -

Enabling Efficient Inference of Convolutional Neural Networks via Approximation

Abstract

UN SDGs

Keywords

Access to Document

Other files and links

Fingerprint

Cite this