Towards a Certified Proof Checker for Deep Neural Network Verification

Remi Desmartin; Omri Isac; Grant Passmore; Kathrin Stark; Ekaterina Komendantskaya; Guy Katz

doi:10.1007/978-3-031-45784-5_13

Towards a Certified Proof Checker for Deep Neural Network Verification

Remi Desmartin, Omri Isac^*, Grant Passmore, Kathrin Stark, Ekaterina Komendantskaya, 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

2 Scopus citations

Abstract

Recent developments in deep neural networks (DNNs) have led to their adoption in safety-critical systems, which in turn has heightened the need for guaranteeing their safety. These safety properties of DNNs can be proven using tools developed by the verification community. However, these tools are themselves prone to implementation bugs and numerical stability problems, which make their reliability questionable. To overcome this, some verifiers produce proofs of their results which can be checked by a trusted checker. In this work, we present a novel implementation of a proof checker for DNN verification. It improves on existing implementations by offering numerical stability and greater verifiability. To achieve this, we leverage two key capabilities of Imandra, an industrial theorem prover: its support for exact real arithmetic and its formal verification infrastructure. So far, we have implemented a proof checker in Imandra, specified its correctness properties and started to verify the checker’s compliance with them. Our ongoing work focuses on completing the formal verification of the checker and further optimising its performance.

Original language	English
Title of host publication	Logic-Based Program Synthesis and Transformation - 33rd International Symposium, LOPSTR 2023, Proceedings
Editors	Robert Glück, Bishoksan Kafle
Publisher	Springer Science and Business Media Deutschland GmbH
Pages	198-209
Number of pages	12
ISBN (Print)	9783031457838
DOIs	https://doi.org/10.1007/978-3-031-45784-5_13
State	Published - 2023
Event	33rd International Symposium on Logic-Based Program Synthesis and Transformation, LOPSTR 2023 - Cascais, Portugal Duration: 23 Oct 2023 → 24 Oct 2023

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	14330 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	33rd International Symposium on Logic-Based Program Synthesis and Transformation, LOPSTR 2023
Country/Territory	Portugal
City	Cascais
Period	23/10/23 → 24/10/23

Bibliographical note

Publisher Copyright:
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023.

Keywords

AI Safety
Deep Neural Network
Formal Verification

Access to Document

10.1007/978-3-031-45784-5_13

Cite this

Desmartin, R., Isac, O., Passmore, G., Stark, K., Komendantskaya, E., & Katz, G. (2023). Towards a Certified Proof Checker for Deep Neural Network Verification. In R. Glück, & B. Kafle (Eds.), Logic-Based Program Synthesis and Transformation - 33rd International Symposium, LOPSTR 2023, Proceedings (pp. 198-209). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 14330 LNCS). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-3-031-45784-5_13

Desmartin, Remi ; Isac, Omri ; Passmore, Grant et al. / Towards a Certified Proof Checker for Deep Neural Network Verification. Logic-Based Program Synthesis and Transformation - 33rd International Symposium, LOPSTR 2023, Proceedings. editor / Robert Glück ; Bishoksan Kafle. Springer Science and Business Media Deutschland GmbH, 2023. pp. 198-209 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{e7018163388e47349683470b7338548d,

title = "Towards a Certified Proof Checker for Deep Neural Network Verification",

abstract = "Recent developments in deep neural networks (DNNs) have led to their adoption in safety-critical systems, which in turn has heightened the need for guaranteeing their safety. These safety properties of DNNs can be proven using tools developed by the verification community. However, these tools are themselves prone to implementation bugs and numerical stability problems, which make their reliability questionable. To overcome this, some verifiers produce proofs of their results which can be checked by a trusted checker. In this work, we present a novel implementation of a proof checker for DNN verification. It improves on existing implementations by offering numerical stability and greater verifiability. To achieve this, we leverage two key capabilities of Imandra, an industrial theorem prover: its support for exact real arithmetic and its formal verification infrastructure. So far, we have implemented a proof checker in Imandra, specified its correctness properties and started to verify the checker{\textquoteright}s compliance with them. Our ongoing work focuses on completing the formal verification of the checker and further optimising its performance.",

keywords = "AI Safety, Deep Neural Network, Formal Verification",

author = "Remi Desmartin and Omri Isac and Grant Passmore and Kathrin Stark and Ekaterina Komendantskaya and Guy Katz",

note = "Publisher Copyright: {\textcopyright} The Author(s), under exclusive license to Springer Nature Switzerland AG 2023.; 33rd International Symposium on Logic-Based Program Synthesis and Transformation, LOPSTR 2023 ; Conference date: 23-10-2023 Through 24-10-2023",

year = "2023",

doi = "10.1007/978-3-031-45784-5_13",

language = "אנגלית",

isbn = "9783031457838",

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

publisher = "Springer Science and Business Media Deutschland GmbH",

pages = "198--209",

editor = "Robert Gl{\"u}ck and Bishoksan Kafle",

booktitle = "Logic-Based Program Synthesis and Transformation - 33rd International Symposium, LOPSTR 2023, Proceedings",

address = "גרמניה",

}

Desmartin, R, Isac, O, Passmore, G, Stark, K, Komendantskaya, E & Katz, G 2023, Towards a Certified Proof Checker for Deep Neural Network Verification. in R Glück & B Kafle (eds), Logic-Based Program Synthesis and Transformation - 33rd International Symposium, LOPSTR 2023, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 14330 LNCS, Springer Science and Business Media Deutschland GmbH, pp. 198-209, 33rd International Symposium on Logic-Based Program Synthesis and Transformation, LOPSTR 2023, Cascais, Portugal, 23/10/23. https://doi.org/10.1007/978-3-031-45784-5_13

Towards a Certified Proof Checker for Deep Neural Network Verification. / Desmartin, Remi; Isac, Omri; Passmore, Grant et al.
Logic-Based Program Synthesis and Transformation - 33rd International Symposium, LOPSTR 2023, Proceedings. ed. / Robert Glück; Bishoksan Kafle. Springer Science and Business Media Deutschland GmbH, 2023. p. 198-209 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 14330 LNCS).

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

TY - GEN

T1 - Towards a Certified Proof Checker for Deep Neural Network Verification

AU - Desmartin, Remi

AU - Isac, Omri

AU - Passmore, Grant

AU - Stark, Kathrin

AU - Komendantskaya, Ekaterina

AU - Katz, Guy

N1 - Publisher Copyright: © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023.

PY - 2023

Y1 - 2023

N2 - Recent developments in deep neural networks (DNNs) have led to their adoption in safety-critical systems, which in turn has heightened the need for guaranteeing their safety. These safety properties of DNNs can be proven using tools developed by the verification community. However, these tools are themselves prone to implementation bugs and numerical stability problems, which make their reliability questionable. To overcome this, some verifiers produce proofs of their results which can be checked by a trusted checker. In this work, we present a novel implementation of a proof checker for DNN verification. It improves on existing implementations by offering numerical stability and greater verifiability. To achieve this, we leverage two key capabilities of Imandra, an industrial theorem prover: its support for exact real arithmetic and its formal verification infrastructure. So far, we have implemented a proof checker in Imandra, specified its correctness properties and started to verify the checker’s compliance with them. Our ongoing work focuses on completing the formal verification of the checker and further optimising its performance.

AB - Recent developments in deep neural networks (DNNs) have led to their adoption in safety-critical systems, which in turn has heightened the need for guaranteeing their safety. These safety properties of DNNs can be proven using tools developed by the verification community. However, these tools are themselves prone to implementation bugs and numerical stability problems, which make their reliability questionable. To overcome this, some verifiers produce proofs of their results which can be checked by a trusted checker. In this work, we present a novel implementation of a proof checker for DNN verification. It improves on existing implementations by offering numerical stability and greater verifiability. To achieve this, we leverage two key capabilities of Imandra, an industrial theorem prover: its support for exact real arithmetic and its formal verification infrastructure. So far, we have implemented a proof checker in Imandra, specified its correctness properties and started to verify the checker’s compliance with them. Our ongoing work focuses on completing the formal verification of the checker and further optimising its performance.

KW - AI Safety

KW - Deep Neural Network

KW - Formal Verification

UR - http://www.scopus.com/inward/record.url?scp=85175806265&partnerID=8YFLogxK

U2 - 10.1007/978-3-031-45784-5_13

DO - 10.1007/978-3-031-45784-5_13

M3 - ???researchoutput.researchoutputtypes.contributiontobookanthology.conference???

AN - SCOPUS:85175806265

SN - 9783031457838

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

SP - 198

EP - 209

BT - Logic-Based Program Synthesis and Transformation - 33rd International Symposium, LOPSTR 2023, Proceedings

A2 - Glück, Robert

A2 - Kafle, Bishoksan

PB - Springer Science and Business Media Deutschland GmbH

T2 - 33rd International Symposium on Logic-Based Program Synthesis and Transformation, LOPSTR 2023

Y2 - 23 October 2023 through 24 October 2023

ER -

Desmartin R, Isac O, Passmore G, Stark K, Komendantskaya E, Katz G. Towards a Certified Proof Checker for Deep Neural Network Verification. In Glück R, Kafle B, editors, Logic-Based Program Synthesis and Transformation - 33rd International Symposium, LOPSTR 2023, Proceedings. Springer Science and Business Media Deutschland GmbH. 2023. p. 198-209. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-031-45784-5_13