Untargeted Code Authorship Evasion with Seq2Seq Transformation

Soohyeon Choi; Rhongho Jang; Dae Hun Nyang; David Mohaisen

doi:10.1007/978-981-97-0669-3_8

Untargeted Code Authorship Evasion with Seq2Seq Transformation

Soohyeon Choi, Rhongho Jang, Dae Hun Nyang, David Mohaisen

Department of Cyber Security

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

Abstract

Code authorship attribution is the problem of identifying authors of programming language codes through the stylistic features in their codes, a topic that recently witnessed significant interest with outstanding performance. In this work, we present SCAE, a code authorship obfuscation technique that leverages a Seq2Seq code transformer called StructCoder. SCAE customizes StructCoder, a system designed initially for function-level code translation from one language to another (e.g., Java to C#), using transfer learning. SCAE improved the efficiency at a slight accuracy degradation compared to existing work. We also reduced the processing time by ≈ 68% while maintaining an 85% transformation success rate and up to 95.77% evasion success rate in the untargeted setting.

Original language	English
Title of host publication	Computational Data and Social Networks - 12th International Conference, CSoNet 2023, Proceedings
Editors	Minh Hoàng Hà, Xingquan Zhu, My T. Thai
Publisher	Springer Science and Business Media Deutschland GmbH
Pages	83-92
Number of pages	10
ISBN (Print)	9789819706686
DOIs	https://doi.org/10.1007/978-981-97-0669-3_8
State	Published - 2024
Event	12th International Conference on Computational Data and Social Networks, CSoNet 2023 - Hanoi, Viet Nam Duration: 11 Dec 2023 → 13 Dec 2023

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	14479 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	12th International Conference on Computational Data and Social Networks, CSoNet 2023
Country/Territory	Viet Nam
City	Hanoi
Period	11/12/23 → 13/12/23

Bibliographical note

Publisher Copyright:
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024.

Keywords

Code Authorship Evasion Attack
Code Authorship Identification
Machine Learning Identification
Program Stylistic Features
Software Forensics

Access to Document

10.1007/978-981-97-0669-3_8

Cite this

Choi, S., Jang, R., Nyang, D. H., & Mohaisen, D. (2024). Untargeted Code Authorship Evasion with Seq2Seq Transformation. In M. H. Hà, X. Zhu, & M. T. Thai (Eds.), Computational Data and Social Networks - 12th International Conference, CSoNet 2023, Proceedings (pp. 83-92). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 14479 LNCS). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-981-97-0669-3_8

Choi, Soohyeon ; Jang, Rhongho ; Nyang, Dae Hun et al. / Untargeted Code Authorship Evasion with Seq2Seq Transformation. Computational Data and Social Networks - 12th International Conference, CSoNet 2023, Proceedings. editor / Minh Hoàng Hà ; Xingquan Zhu ; My T. Thai. Springer Science and Business Media Deutschland GmbH, 2024. pp. 83-92 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{e3f97c243d0f4714be581ca0fe419031,

title = "Untargeted Code Authorship Evasion with Seq2Seq Transformation",

abstract = "Code authorship attribution is the problem of identifying authors of programming language codes through the stylistic features in their codes, a topic that recently witnessed significant interest with outstanding performance. In this work, we present SCAE, a code authorship obfuscation technique that leverages a Seq2Seq code transformer called StructCoder. SCAE customizes StructCoder, a system designed initially for function-level code translation from one language to another (e.g., Java to C#), using transfer learning. SCAE improved the efficiency at a slight accuracy degradation compared to existing work. We also reduced the processing time by ≈ 68% while maintaining an 85% transformation success rate and up to 95.77% evasion success rate in the untargeted setting.",

keywords = "Code Authorship Evasion Attack, Code Authorship Identification, Machine Learning Identification, Program Stylistic Features, Software Forensics",

author = "Soohyeon Choi and Rhongho Jang and Nyang, {Dae Hun} and David Mohaisen",

note = "Publisher Copyright: {\textcopyright} The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024.; 12th International Conference on Computational Data and Social Networks, CSoNet 2023 ; Conference date: 11-12-2023 Through 13-12-2023",

year = "2024",

doi = "10.1007/978-981-97-0669-3_8",

language = "English",

isbn = "9789819706686",

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 = "83--92",

editor = "H{\`a}, {Minh Ho{\`a}ng} and Xingquan Zhu and Thai, {My T.}",

booktitle = "Computational Data and Social Networks - 12th International Conference, CSoNet 2023, Proceedings",

}

Choi, S, Jang, R, Nyang, DH & Mohaisen, D 2024, Untargeted Code Authorship Evasion with Seq2Seq Transformation. in MH Hà, X Zhu & MT Thai (eds), Computational Data and Social Networks - 12th International Conference, CSoNet 2023, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 14479 LNCS, Springer Science and Business Media Deutschland GmbH, pp. 83-92, 12th International Conference on Computational Data and Social Networks, CSoNet 2023, Hanoi, Viet Nam, 11/12/23. https://doi.org/10.1007/978-981-97-0669-3_8

Untargeted Code Authorship Evasion with Seq2Seq Transformation. / Choi, Soohyeon; Jang, Rhongho; Nyang, Dae Hun et al.
Computational Data and Social Networks - 12th International Conference, CSoNet 2023, Proceedings. ed. / Minh Hoàng Hà; Xingquan Zhu; My T. Thai. Springer Science and Business Media Deutschland GmbH, 2024. p. 83-92 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 14479 LNCS).

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

TY - GEN

T1 - Untargeted Code Authorship Evasion with Seq2Seq Transformation

AU - Choi, Soohyeon

AU - Jang, Rhongho

AU - Nyang, Dae Hun

AU - Mohaisen, David

N1 - Publisher Copyright: © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024.

PY - 2024

Y1 - 2024

N2 - Code authorship attribution is the problem of identifying authors of programming language codes through the stylistic features in their codes, a topic that recently witnessed significant interest with outstanding performance. In this work, we present SCAE, a code authorship obfuscation technique that leverages a Seq2Seq code transformer called StructCoder. SCAE customizes StructCoder, a system designed initially for function-level code translation from one language to another (e.g., Java to C#), using transfer learning. SCAE improved the efficiency at a slight accuracy degradation compared to existing work. We also reduced the processing time by ≈ 68% while maintaining an 85% transformation success rate and up to 95.77% evasion success rate in the untargeted setting.

AB - Code authorship attribution is the problem of identifying authors of programming language codes through the stylistic features in their codes, a topic that recently witnessed significant interest with outstanding performance. In this work, we present SCAE, a code authorship obfuscation technique that leverages a Seq2Seq code transformer called StructCoder. SCAE customizes StructCoder, a system designed initially for function-level code translation from one language to another (e.g., Java to C#), using transfer learning. SCAE improved the efficiency at a slight accuracy degradation compared to existing work. We also reduced the processing time by ≈ 68% while maintaining an 85% transformation success rate and up to 95.77% evasion success rate in the untargeted setting.

KW - Code Authorship Evasion Attack

KW - Code Authorship Identification

KW - Machine Learning Identification

KW - Program Stylistic Features

KW - Software Forensics

UR - http://www.scopus.com/inward/record.url?scp=85187647841&partnerID=8YFLogxK

U2 - 10.1007/978-981-97-0669-3_8

DO - 10.1007/978-981-97-0669-3_8

M3 - Conference contribution

AN - SCOPUS:85187647841

SN - 9789819706686

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

SP - 83

EP - 92

BT - Computational Data and Social Networks - 12th International Conference, CSoNet 2023, Proceedings

A2 - Hà, Minh Hoàng

A2 - Zhu, Xingquan

A2 - Thai, My T.

PB - Springer Science and Business Media Deutschland GmbH

T2 - 12th International Conference on Computational Data and Social Networks, CSoNet 2023

Y2 - 11 December 2023 through 13 December 2023

ER -

Choi S, Jang R, Nyang DH, Mohaisen D. Untargeted Code Authorship Evasion with Seq2Seq Transformation. In Hà MH, Zhu X, Thai MT, editors, Computational Data and Social Networks - 12th International Conference, CSoNet 2023, Proceedings. Springer Science and Business Media Deutschland GmbH. 2024. p. 83-92. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-981-97-0669-3_8