Keivan Rezaei

About Me

I am a second-year Ph.D. student at the University of Maryland, advised by Prof. Feizi and Prof. Hajiaghayi.

My research centers around Deep Learning Robustness and Interpretability, with a keen interest in comprehending what models learn, how they utilize their knowledge for predictions, and understanding their successes and failures. I strive to represent these concepts in easily understandable terms, often using languages.

Research Interests

• Deep Learning Robustness and Interpretability: data poisoning, vision-language models, human-understandable explainability.
• Game Theory: mechanism design, quality of equilibria.

News

• [Jan 2024] Our papers “PRIME: Prioritizing Interpretability in Failure Mode Extraction” and “Robustness of AI-Image Detectors: Fundamental Limits and Practical Attacks” are accepted to ICLR 2024.
• [Dec 2023] Our paper “Regret Analysis of Repeated Delegated Choice” is accepted to AAAI 2024.
• [Nov 2023] We posted our draft “Online Advertisements with LLMs: Opportunities and Challenges” on arXiv.

• [Oct. 2023] Our work “Robustness of AI-Image Detectors: Fundamental Limits and Practical Attacks” was covered by Register, Wired, and Arstechnica.

• [May. 2023] Our team, UMD RED, ranked 3-rd in ICPC NAC 2023 and proceeded to ICPC World Finals 2023.
• [May. 2023] We submitted our manuscript “Regret Analysis of Repeated Delegated Choice” to arXiv.
• [May. 2023] Our paper “Delegating to Multiple Agents” is accepted to EC 2023.
• [Apr. 2023] Our paper “Run-off Election: Improved Provable Defense against Data Poisoning Attacks” is accepted to ICML 2023.
• [Apr. 2023] Our paper “Text-To-Concept (and Back) via Cross-Model Alignment” is accepted to ICML 2023.
• [Feb. 2023] Our team, UMD RED, is qualified to compete in ICPC NAC 2023 at UCF.

Publications

Services

Conference Reviewers

• Conference on Neural Information Processing Systems (NeurIPS) 2023
• International Conference on Learning Representations (ICLR) 2024

Powered by Jekyll and Minimal Light theme.