David Shriver
Contact
Email: davidshriver@outlook.com
Website: https://dlshriver.com
Research Interests
I am a machine learning research scientist in the AI Division of the Software Engineering Institute (SEI) at Carnegie Mellon University (CMU). My research interests are in ensuring correctness of machine learning systems. In particular, I am interested in analysis, testing, and verification of neural networks, especially in the context of systems with a high cost of failure, such as safety-critical and autonomous systems.
Education
Ph.D., Computer Science
•
Dec 2022
•
University of Virginia
Advised by:
Matt Dwyer and Sebastian Elbaum
Thesis: Increasing the Applicability of Verification Tools for Neural Networks
M.S., Computer Science
•
May 2018
•
University of Nebraska-Lincoln
Advised by:
Sebastian Elbaum
Thesis: Assessing the Quality and Stability of Recommender Systems
B.S., Computer Engineering
•
May 2016
•
University of Nebraska-Lincoln
Positions Held
Machine Learning Research Scientist
•
Software Engineering Institute (SEI)
January 2023 — Present
Leads small teams of researchers, engineers, and software developers on projects in the Secure AI Lab on topics related to identifying, understanding, and defending against AI model vulnerabilities.
Research Assistant
•
Department of Computer Science, University of Virginia
August 2018 — December 2022
Performed research on techniques for increasing the applicability of state-of-the art formal verification tools for neural networks to improve the safety and correctness of systems with high costs of failure, such as autonomous vehicles.
Research Intern
•
Langley Research Center, NASA
June 2021 — August 2021
Proposed and led a short term project to develop a method for transforming temporal behavioral properties of neural networks to local robustness properties to enable application of a plethora of available tools and investigated the feasibility of this transformation in the context of neural networks trained for aircraft collision avoidance.
Research Assistant
•
Department of Computer Science and Engineering, University of Nebraska-Lincoln
March 2014 — July 2018
Developed approaches for assessing the quality and stability of recommender systems.
Publications
Concept-ROT: Poisoning Concepts in Large Language Models with Model Editing
Keltin Grimes,
Marco Christiani,
,
Marissa Connor
arXiv preprint arXiv:2412.13341
—
2024
Keltin Grimes, Marco Christiani, David Shriver, Marissa Connor. 2024. Concept-ROT: Poisoning Concepts in Large Language Models with Model Editing. arXiv:2412.13341. https://arxiv.org/abs/2412.13341
Deeper Notions of Correctness in Image-Based DNNs: Lifting Properties from Pixel to Entities
Felipe Toledo,
,
Sebastian Elbaum,
Matthew B. Dwyer
Foundations of Software Engineering
—
2023
[Paper]
Felipe Toledo, David Shriver, Sebastian Elbaum, Matthew B. Dwyer. 2023. Deeper Notions of Correctness in Image-Based DNNs: Lifting Properties from Pixel to Entities. In Proceedings of the 31st ACM Joint European Software Engineering Conference and Symposium on the Foundations of Software Engineering (ESEC/FSE 2023). Association for Computing Machinery, New York, NY, USA, 2122–2126. https://doi.org/10.1145/3611643.3613079
DeepManeuver: Adversarial Test Generation for Trajectory Manipulation of Autonomous Vehicles
Meriel von Stein,
,
Sebastian Elbaum
IEEE Transactions on Software Engineering
—
2023
Meriel von Stein, David Shriver, Sebastian Elbaum. 2023. DeepManeuver: Adversarial Test Generation for Trajectory Manipulation of Autonomous Vehicles. In IEEE Transactions on Software Engineering, vol. 49, no. 10, pp. 4496-4509. https://doi.org/10.1109/TSE.2023.3301443
Increasing the Applicability of Verification Tools for Neural Networks
University of Virginia
—
2022
[Paper]
David Shriver. Increasing the Applicability of Verification Tools for Neural Networks. PhD Thesis. University of Virginia. 2022.
Distribution Models for Falsification and Verification of DNNs
Felipe Toledo,
,
Sebastian Elbaum,
Matthew B. Dwyer
International Conference on Automated Software Engineering (ASE)
—
2021
Felipe Toledo, David Shriver, Sebastian Elbaum, Matthew B. Dwyer. 2021. Distribution Models for Falsification and Verification of DNNs. In 2021 36th IEEE/ACM International Conference on Automated Software Engineering (ASE). 317-329. https://doi.org/10.1109/ASE51524.2021.9678590
DNNV: A Framework for Deep Neural Network Verification
,
Sebastian Elbaum,
Matthew B. Dwyer
Computer Aided Verification
—
2021
David Shriver, Sebastian Elbaum, Matthew B. Dwyer. 2021. DNNV: A Framework for Deep Neural Network Verification. In Computer Aided Verification. 137--150. https://doi.org/10.1007/978-3-030-81685-8_6
Reducing DNN Properties to Enable Falsification with Adversarial Attacks
,
Sebastian Elbaum,
Matthew B. Dwyer
International Conference on Software Engineering (ICSE)
—
2021
David Shriver, Sebastian Elbaum, Matthew B. Dwyer. 2021. Reducing DNN Properties to Enable Falsification with Adversarial Attacks. In 2021 IEEE/ACM 43rd International Conference on Software Engineering (ICSE). 275-287. https://doi.org/10.1109/ICSE43902.2021.00036
Systematic Generation of Diverse Benchmarks for DNN Verification
Dong Xu,
,
Matthew B. Dwyer,
Sebastian Elbaum
Computer Aided Verification
—
2020
[Paper]
Dong Xu, David Shriver, Matthew B. Dwyer, Sebastian Elbaum. 2020. Systematic Generation of Diverse Benchmarks for DNN Verification. In Computer Aided Verification - 32nd International Conference, CAV 2020, Los Angeles, CA, USA, July 21-24, 2020, Proceedings, Part I. 97-121. https://doi.org/10.1007/978-3-030-53288-8_5
Poster: Differencing Neural Networks
,
Sebastian Elbaum,
Matthew B. Dwyer
University of Virginia CS Research Symposium
—
2019
[Poster]
David Shriver, Sebastian Elbaum, Matthew B. Dwyer. 2019. Differencing Neural Networks. In the 2019 University of Virginia Computer Science Research Symposium.
Refactoring Neural Networks for Verification
,
Dong Xu,
Sebastian Elbaum,
Matthew B. Dwyer
arXiv preprint arXiv:1908.08026
—
2019
[Paper]
David Shriver, Dong Xu, Sebastian Elbaum, Matthew B. Dwyer. 2019. Refactoring Neural Networks for Verification. arXiv:1908.08026. https://arxiv.org/abs/1908.08026
Evaluating Recommender System Stability with Influence-Guided Fuzzing
,
Sebastian Elbaum,
Matthew B. Dwyer,
David S. Rosenblum
AAAI Conference on Artificial Intelligence
—
2019
David Shriver, Sebastian Elbaum, Matthew B. Dwyer, David S. Rosenblum. 2019. Evaluating Recommender System Stability with Influence-Guided Fuzzing. In The Thirty-Third AAAI Conference on Artificial Intelligence, AAAI 2019, The Thirty-First Innovative Applications of Artificial Intelligence Conference, IAAI 2019, The Ninth AAAI Symposium on Educational Advances in Artificial Intelligence, EAAI 2019, Honolulu, Hawaii, USA, January 27 - February 1, 2019. 4934-4942. https://doi.org/10.1609/aaai.v33i01.33014934
Toward the development of richer properties for recommender systems
International Conference on Software Engineering: Companion Proceedings
—
2018
David Shriver. 2018. Toward the development of richer properties for recommender systems. In Proceedings of the 40th International Conference on Software Engineering: Companion Proceedings, ICSE 2018, Gothenburg, Sweden, May 27 - June 03, 2018. 173-174. https://doi.acm.org/10.1145/3183440.3195082
Assessing the Quality and Stability of Recommender Systems
University of Nebraska-Lincoln
—
2018
[Paper]
David Shriver. Assessing the Quality and Stability of Recommender Systems. MS Thesis, University of Nebraska-Lincoln, 2018.
At the End of Synthesis: Narrowing Program Candidates
,
Sebastian G. Elbaum,
Kathryn T. Stolee
International Conference on Software Engineering: New Ideas and Emerging Technologies Results Track
—
2017
[Paper]
David Shriver, Sebastian G. Elbaum, Kathryn T. Stolee. 2017. At the End of Synthesis: Narrowing Program Candidates. In 39th IEEE/ACM International Conference on Software Engineering: New Ideas and Emerging Technologies Results Track, ICSE-NIER 2017, Buenos Aires, Argentina, May 20-28, 2017. 19-22. https://doi.org/10.1109/ICSE-NIER.2017.7
Tools and Artifacts
dnnf
:Implements a reduction to enable the application of falsification tools, such as adversarial attacks, to a more general set of behavioral properties of neural networks.
https://github.com/dlshriver/dnnf[Github] [Documentation] [Video] [Docker]
dnnv-benchmarks
:A large collection of DNN verification benchmarks, specified in DNNP and ONNX for use with DNNV, DNNF and their supported verifiers and falsifiers.
https://github.com/dlshriver/dnnv-benchmarks[Github]
dnnv
:Introduces a standard network and property specification formats and implements network simplifications and property reductions, facilitating verifier execution, comparison, and artifact re-use.
https://github.com/dlshriver/dnnv[Github] [Documentation] [Video] [Docker]
Awards and Honors
• ACM SIGSOFT Outstanding Doctoral Dissertation, March 2023
• John A. Stankovic Outstanding Graduate Research, May 2022
• University of Nebraska-Lincoln, Highest Distinction, May 2016
• University of Nebraska-Lincoln, Computer Engineering Outstanding Undergraduate Senior, May 2016
Service
• ICSE 2025: Program Committee Member
• TOSEM 2024: Reviewer
• TOSEM 2023: Reviewer
• ASE 2022: Committee Member in Program Committee within Artifact Evaluation-track
• ISSTA 2021: Committee Member in Artifact Evaluation Committee within the Artifact Evaluation-track
• ISSTA 2021: Co-reviewer
• ICSE 2020: Co-reviewer