Research

Paper

TESTING February 18, 2026

Exact Certification of Data-Poisoning Attacks Using Mixed-Integer Programming

Authors

Philip Sosnin, Jodie Knapp, Fraser Kennedy, Josh Collyer, Calvin Tsay

Abstract

This work introduces a verification framework that provides both sound and complete guarantees for data poisoning attacks during neural network training. We formulate adversarial data manipulation, model training, and test-time evaluation in a single mixed-integer quadratic programming (MIQCP) problem. Finding the global optimum of the proposed formulation provably yields worst-case poisoning attacks, while simultaneously bounding the effectiveness of all possible attacks on the given training pipeline. Our framework encodes both the gradient-based training dynamics and model evaluation at test time, enabling the first exact certification of training-time robustness. Experimental evaluation on small models confirms that our approach delivers a complete characterization of robustness against data poisoning.

Metadata

arXiv ID: 2602.16944
Provider: ARXIV
Primary Category: cs.LG
Published: 2026-02-18
Fetched: 2026-02-21 18:51

Related papers

Raw Data (Debug)
{
  "raw_xml": "<entry>\n    <id>http://arxiv.org/abs/2602.16944v1</id>\n    <title>Exact Certification of Data-Poisoning Attacks Using Mixed-Integer Programming</title>\n    <updated>2026-02-18T23:18:45Z</updated>\n    <link href='https://arxiv.org/abs/2602.16944v1' rel='alternate' type='text/html'/>\n    <link href='https://arxiv.org/pdf/2602.16944v1' rel='related' title='pdf' type='application/pdf'/>\n    <summary>This work introduces a verification framework that provides both sound and complete guarantees for data poisoning attacks during neural network training. We formulate adversarial data manipulation, model training, and test-time evaluation in a single mixed-integer quadratic programming (MIQCP) problem. Finding the global optimum of the proposed formulation provably yields worst-case poisoning attacks, while simultaneously bounding the effectiveness of all possible attacks on the given training pipeline. Our framework encodes both the gradient-based training dynamics and model evaluation at test time, enabling the first exact certification of training-time robustness. Experimental evaluation on small models confirms that our approach delivers a complete characterization of robustness against data poisoning.</summary>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.LG'/>\n    <published>2026-02-18T23:18:45Z</published>\n    <arxiv:comment>Accepted to the 23rd International Conference on the Integration of Constraint Programming, Artificial Intelligence, and Operations Research (CPAIOR)</arxiv:comment>\n    <arxiv:primary_category term='cs.LG'/>\n    <author>\n      <name>Philip Sosnin</name>\n    </author>\n    <author>\n      <name>Jodie Knapp</name>\n    </author>\n    <author>\n      <name>Fraser Kennedy</name>\n    </author>\n    <author>\n      <name>Josh Collyer</name>\n    </author>\n    <author>\n      <name>Calvin Tsay</name>\n    </author>\n  </entry>"
}