Simplifying Neural Networks Using Formal Verification

Sumathi Gokulanathan; Alexander Feldsher; Adi Malca; Clark Barrett; Guy Katz

doi:10.1007/978-3-030-55754-6_5

Simplifying Neural Networks Using Formal Verification

Sumathi Gokulanathan, Alexander Feldsher, Adi Malca, Clark Barrett, Guy Katz^*

^*Corresponding author for this work

The Rachel and Selim Benin School of Engineering and Computer Science

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

19 Scopus citations

Abstract

Deep neural network (DNN) verification is an emerging field, with diverse verification engines quickly becoming available. Demonstrating the effectiveness of these engines on real-world DNNs is an important step towards their wider adoption. We present a tool that can leverage existing verification engines in performing a novel application: neural network simplification, through the reduction of the size of a DNN without harming its accuracy. We report on the work-flow of the simplification process, and demonstrate its potential significance and applicability on a family of real-world DNNs for aircraft collision avoidance, whose sizes we were able to reduce by as much as 10%.

Original language	American English
Title of host publication	NASA Formal Methods - 12th International Symposium, NFM 2020, Proceedings
Editors	Ritchie Lee, Susmit Jha, Anastasia Mavridou
Publisher	Springer
Pages	85-93
Number of pages	9
ISBN (Print)	9783030557539
DOIs	https://doi.org/10.1007/978-3-030-55754-6_5
State	Published - 2020
Event	12th International Symposium on NASA Formal Methods, NFM 2020 - Moffett Field, United States Duration: 11 May 2020 → 15 May 2020

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	12229 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	12th International Symposium on NASA Formal Methods, NFM 2020
Country/Territory	United States
City	Moffett Field
Period	11/05/20 → 15/05/20

Bibliographical note

Funding Information:
Acknowledgements. This project was partially supported by grants from the Binational Science Foundation (2017662), the Israel Science Foundation (683/18), and the National Science Foundation (1814369).

Publisher Copyright:
© 2020, Springer Nature Switzerland AG.

Keywords

Deep neural networks
Marabou
Simplification
Verification

Access to Document

10.1007/978-3-030-55754-6_5

Cite this

Gokulanathan, S., Feldsher, A., Malca, A., Barrett, C., & Katz, G. (2020). Simplifying Neural Networks Using Formal Verification. In R. Lee, S. Jha, & A. Mavridou (Eds.), NASA Formal Methods - 12th International Symposium, NFM 2020, Proceedings (pp. 85-93). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 12229 LNCS). Springer. https://doi.org/10.1007/978-3-030-55754-6_5

Gokulanathan, Sumathi ; Feldsher, Alexander ; Malca, Adi et al. / Simplifying Neural Networks Using Formal Verification. NASA Formal Methods - 12th International Symposium, NFM 2020, Proceedings. editor / Ritchie Lee ; Susmit Jha ; Anastasia Mavridou. Springer, 2020. pp. 85-93 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{b53007541bbf4a7fbc0bb475af181e3a,

title = "Simplifying Neural Networks Using Formal Verification",

abstract = "Deep neural network (DNN) verification is an emerging field, with diverse verification engines quickly becoming available. Demonstrating the effectiveness of these engines on real-world DNNs is an important step towards their wider adoption. We present a tool that can leverage existing verification engines in performing a novel application: neural network simplification, through the reduction of the size of a DNN without harming its accuracy. We report on the work-flow of the simplification process, and demonstrate its potential significance and applicability on a family of real-world DNNs for aircraft collision avoidance, whose sizes we were able to reduce by as much as 10%.",

keywords = "Deep neural networks, Marabou, Simplification, Verification",

author = "Sumathi Gokulanathan and Alexander Feldsher and Adi Malca and Clark Barrett and Guy Katz",

note = "Funding Information: Acknowledgements. This project was partially supported by grants from the Binational Science Foundation (2017662), the Israel Science Foundation (683/18), and the National Science Foundation (1814369). Publisher Copyright: {\textcopyright} 2020, Springer Nature Switzerland AG.; 12th International Symposium on NASA Formal Methods, NFM 2020 ; Conference date: 11-05-2020 Through 15-05-2020",

year = "2020",

doi = "10.1007/978-3-030-55754-6_5",

language = "American English",

isbn = "9783030557539",

series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

publisher = "Springer",

pages = "85--93",

editor = "Ritchie Lee and Susmit Jha and Anastasia Mavridou",

booktitle = "NASA Formal Methods - 12th International Symposium, NFM 2020, Proceedings",

}

Gokulanathan, S, Feldsher, A, Malca, A, Barrett, C & Katz, G 2020, Simplifying Neural Networks Using Formal Verification. in R Lee, S Jha & A Mavridou (eds), NASA Formal Methods - 12th International Symposium, NFM 2020, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 12229 LNCS, Springer, pp. 85-93, 12th International Symposium on NASA Formal Methods, NFM 2020, Moffett Field, United States, 11/05/20. https://doi.org/10.1007/978-3-030-55754-6_5

Simplifying Neural Networks Using Formal Verification. / Gokulanathan, Sumathi; Feldsher, Alexander; Malca, Adi et al.
NASA Formal Methods - 12th International Symposium, NFM 2020, Proceedings. ed. / Ritchie Lee; Susmit Jha; Anastasia Mavridou. Springer, 2020. p. 85-93 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 12229 LNCS).

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

TY - GEN

T1 - Simplifying Neural Networks Using Formal Verification

AU - Gokulanathan, Sumathi

AU - Feldsher, Alexander

AU - Malca, Adi

AU - Barrett, Clark

AU - Katz, Guy

N1 - Funding Information: Acknowledgements. This project was partially supported by grants from the Binational Science Foundation (2017662), the Israel Science Foundation (683/18), and the National Science Foundation (1814369). Publisher Copyright: © 2020, Springer Nature Switzerland AG.

PY - 2020

Y1 - 2020

N2 - Deep neural network (DNN) verification is an emerging field, with diverse verification engines quickly becoming available. Demonstrating the effectiveness of these engines on real-world DNNs is an important step towards their wider adoption. We present a tool that can leverage existing verification engines in performing a novel application: neural network simplification, through the reduction of the size of a DNN without harming its accuracy. We report on the work-flow of the simplification process, and demonstrate its potential significance and applicability on a family of real-world DNNs for aircraft collision avoidance, whose sizes we were able to reduce by as much as 10%.

AB - Deep neural network (DNN) verification is an emerging field, with diverse verification engines quickly becoming available. Demonstrating the effectiveness of these engines on real-world DNNs is an important step towards their wider adoption. We present a tool that can leverage existing verification engines in performing a novel application: neural network simplification, through the reduction of the size of a DNN without harming its accuracy. We report on the work-flow of the simplification process, and demonstrate its potential significance and applicability on a family of real-world DNNs for aircraft collision avoidance, whose sizes we were able to reduce by as much as 10%.

KW - Deep neural networks

KW - Marabou

KW - Simplification

KW - Verification

UR - http://www.scopus.com/inward/record.url?scp=85089722312&partnerID=8YFLogxK

U2 - 10.1007/978-3-030-55754-6_5

DO - 10.1007/978-3-030-55754-6_5

M3 - Conference contribution

AN - SCOPUS:85089722312

SN - 9783030557539

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 85

EP - 93

BT - NASA Formal Methods - 12th International Symposium, NFM 2020, Proceedings

A2 - Lee, Ritchie

A2 - Jha, Susmit

A2 - Mavridou, Anastasia

PB - Springer

T2 - 12th International Symposium on NASA Formal Methods, NFM 2020

Y2 - 11 May 2020 through 15 May 2020

ER -

Gokulanathan S, Feldsher A, Malca A, Barrett C, Katz G. Simplifying Neural Networks Using Formal Verification. In Lee R, Jha S, Mavridou A, editors, NASA Formal Methods - 12th International Symposium, NFM 2020, Proceedings. Springer. 2020. p. 85-93. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-030-55754-6_5

Simplifying Neural Networks Using Formal Verification

Abstract

Publication series

Conference

Bibliographical note

Keywords

Access to Document

Other files and links

Fingerprint

Cite this