Counterexample Guided Neural Network Quantization Refinement

Joao Batista P.  Matos Jr; Eddie B.  de Lima Filho; Iury  Bessa; Edoardo Manino; Xidan Song; Lucas C. Cordeiro

doi:10.1109/TCAD.2023.3335313

Counterexample Guided Neural Network Quantization Refinement

Joao Batista P. Matos Jr, Eddie B. de Lima Filho, Iury Bessa, Edoardo Manino, Xidan Song, Lucas C. Cordeiro

Research output: Contribution to journal › Article › peer-review

30 Downloads (Pure)

Abstract

Deploying Neural networks (NNs) in low-resource domains is challenging because of their high computing, memory, and power requirements. For this reason, NNs are often quantized before deployment, but such an approach degrades their
accuracy. Thus, we propose the counterexample guided neural network quantization refinement (CEG4N) framework, which combines search-based quantization and equivalence checking. The former minimizes computational requirements, while the latter guarantees that the behavior of an NN does not change after quantization. We evaluate CEG4N on a diverse set of benchmarks, including large and small NNs. Our technique successfully quantizes the networks in the chosen evaluation set, while producing models with up to 163% better accuracy than state-of-the-art techniques.

Original language	English
Journal	IEEE Transactions on Computer - Aided Design of Integrated Circuits and Systems
DOIs	https://doi.org/10.1109/TCAD.2023.3335313
Publication status	Published - 21 Nov 2023

Keywords

cs.LG
cs.AI
cs.SE

Access to Document

10.1109/TCAD.2023.3335313Licence: CC BY

1 Finished
1 Active

EnnCore: End-to-End Conceptual Guarding of Neural Architectures
Cordeiro, L. (PI), Brown, G. (CoI), Freitas, A. (CoI), Luján, M. (CoI) & Mustafa, M. (CoI)
1/02/21 → 31/12/25
Project: Research
SCorCH: Secure Code for Capability Hardware
Reger, G. (PI), Cordeiro, L. (CoI), Korovin, K. (CoI), Mustafa, M. (CoI) & Olivier, P. (CoI)
1/07/20 → 31/12/23
Project: Research

Cite this

@article{bfd5810ad8494c818f5b1c27161eb827,

title = "Counterexample Guided Neural Network Quantization Refinement",

abstract = "Deploying Neural networks (NNs) in low-resource domains is challenging because of their high computing, memory, and power requirements. For this reason, NNs are often quantized before deployment, but such an approach degrades theiraccuracy. Thus, we propose the counterexample guided neural network quantization refinement (CEG4N) framework, which combines search-based quantization and equivalence checking. The former minimizes computational requirements, while the latter guarantees that the behavior of an NN does not change after quantization. We evaluate CEG4N on a diverse set of benchmarks, including large and small NNs. Our technique successfully quantizes the networks in the chosen evaluation set, while producing models with up to 163% better accuracy than state-of-the-art techniques.",

keywords = "cs.LG, cs.AI, cs.SE",

author = "{Matos Jr}, {Joao Batista P.} and {de Lima Filho}, {Eddie B.} and Iury Bessa and Edoardo Manino and Xidan Song and Cordeiro, {Lucas C.}",

year = "2023",

month = nov,

day = "21",

doi = "10.1109/TCAD.2023.3335313",

language = "English",

journal = "IEEE Transactions on Computer - Aided Design of Integrated Circuits and Systems ",

issn = "0278-0070",

publisher = "IEEE",

}

TY - JOUR

T1 - Counterexample Guided Neural Network Quantization Refinement

AU - Matos Jr, Joao Batista P.

AU - de Lima Filho, Eddie B.

AU - Bessa, Iury

AU - Manino, Edoardo

AU - Song, Xidan

AU - Cordeiro, Lucas C.

PY - 2023/11/21

Y1 - 2023/11/21

N2 - Deploying Neural networks (NNs) in low-resource domains is challenging because of their high computing, memory, and power requirements. For this reason, NNs are often quantized before deployment, but such an approach degrades theiraccuracy. Thus, we propose the counterexample guided neural network quantization refinement (CEG4N) framework, which combines search-based quantization and equivalence checking. The former minimizes computational requirements, while the latter guarantees that the behavior of an NN does not change after quantization. We evaluate CEG4N on a diverse set of benchmarks, including large and small NNs. Our technique successfully quantizes the networks in the chosen evaluation set, while producing models with up to 163% better accuracy than state-of-the-art techniques.

AB - Deploying Neural networks (NNs) in low-resource domains is challenging because of their high computing, memory, and power requirements. For this reason, NNs are often quantized before deployment, but such an approach degrades theiraccuracy. Thus, we propose the counterexample guided neural network quantization refinement (CEG4N) framework, which combines search-based quantization and equivalence checking. The former minimizes computational requirements, while the latter guarantees that the behavior of an NN does not change after quantization. We evaluate CEG4N on a diverse set of benchmarks, including large and small NNs. Our technique successfully quantizes the networks in the chosen evaluation set, while producing models with up to 163% better accuracy than state-of-the-art techniques.

KW - cs.LG

KW - cs.AI

KW - cs.SE

U2 - 10.1109/TCAD.2023.3335313

DO - 10.1109/TCAD.2023.3335313

M3 - Article

SN - 0278-0070

JO - IEEE Transactions on Computer - Aided Design of Integrated Circuits and Systems

JF - IEEE Transactions on Computer - Aided Design of Integrated Circuits and Systems

ER -

Counterexample Guided Neural Network Quantization Refinement

Abstract

Keywords

Access to Document

Fingerprint

Projects

EnnCore: End-to-End Conceptual Guarding of Neural Architectures

SCorCH: Secure Code for Capability Hardware

Cite this