Certified Robustness and Training
Sparse SOCP verification, the CORE verification cascade, and conic-regularized certified robust training.
Posts by Collection
portfolio
Sequential Distribution-Shift Detection — DI-SCUSUM
Training-free detection using closed-form diffusion-integrated scores, with explicit false-alarm and detection-delay evaluation.
ParagraphAI Experience
Product experimentation and AI-agent workflows are described under ParagraphAI in Experience.
publications
Using Overlapped Resonators in Wireless Power Transfer for Uniform Electromagnetic Field and Removing Blank Spots in Free Moving Applications
Electronics 11(8), 1204, 2022
Overlapping transmitter-coil arrays reduce power gaps as a receiver moves across a wireless-power surface.
Recommended citation: Saeideh Pahlavan, Mostafa Shooshtari, Mohammadreza Maleki, and Shahin Jafarabadi Ashtiani. "Using Overlapped Resonators in Wireless Power Transfer for Uniform Electromagnetic Field and Removing Blank Spots in Free Moving Applications." Electronics 11, no. 8: 1204, 2022.
Download Paper
Cascading Robustness Verification: Toward Efficient Model-Agnostic Certification
IEEE SaTML 2026, 2026
A staged, model-agnostic verification framework that applies stronger certifiers only when cheaper methods cannot certify an input.
Recommended citation: Mohammadreza Maleki, Rushendra Sidibomma, Arman Adibi, and Reza Samavi. "Cascading Robustness Verification: Toward Efficient Model-Agnostic Certification." IEEE Conference on Secure and Trustworthy Machine Learning (SaTML), 2026.
Download Paper
CEAR: Certified Ensemble Adversarial Robustness in DNNs
Canadian AI 2026 · PMLR 318:624–635, 2026
An ensemble method combining noise-trained networks, voting rules, and randomized-smoothing certificates.
Recommended citation: Daniel Sadig, Mohammadreza Maleki, Hamed Karimi, and Reza Samavi. "CEAR: Certified Ensemble Adversarial Robustness in DNNs." Proceedings of the 39th Canadian Conference on Artificial Intelligence, PMLR 318:624–635, 2026.
Download Paper
talks
Rethinking the Tightness–Efficiency Trade-off in Certified Robustness
Published:
Presented in May 2026. Presentation on the computational and certification trade-offs in neural-network robustness verification.
CEAR: Certified Ensemble Adversarial Robustness in DNNs
Published:
Presented CEAR: Certified Ensemble Adversarial Robustness in DNNs at the Canadian Conference on Artificial Intelligence, held at Simon Fraser University in May 2026. The paper develops an ensemble approach that combines diverse noise-trained networks, voting rules, and randomized smoothing to improve certified adversarial robustness.