Mehrdad Saberi

About Me

I am a CS Ph.D. student at University of Maryland since Spring 2023, advised by Prof. Soheil Feizi. My research mainly revolves around Foundation and Multimodal Models, covering topics such as image generation, vision-language models, and interpretability and robustness of machine learning methods.

Research Interests

• Generative AI: diffusion models, multimodal models, large language models.
• Trustworthy AI: adversarial robustness, deepfake detection.
• AI Interpretability: spurious features, failure-mode detection.

News

• [Aug. 2024] I’ve been honored to be part of the organizer team for the Erasing the Invisible: A Stress-Test Challenge for Image Watermarks competition at Neurips 2024.
• [May. 2024] Check out our new work on the risks and opportunities of modern generative AI.
• [Mar. 2024] I had the opportunity to present our work “Robustness of AI-Image Detectors: Fundamental Limits and Practical Attacks” in PrivacyCon 2024 and AGI Leap Summit. (Video Link)
• [Jan. 2024] Two of our papers, “Robustness of AI-Image Detectors: Fundamental Limits and Practical Attacks” and “PRIME: Prioritizing Interpretability in Failure Mode Extraction” got accepted into ICLR 2024 Conference.
• [Oct. 2023] Our work “Robustness of AI-Image Detectors: Fundamental Limits and Practical Attacks” was covered in news articles such as Wired, The Register, and ArtTechnica.
• [Sep. 2023] Our paper “ZeroGrad: Costless conscious remedies for catastrophic overfitting in the FGSM adversarial training” was published by Elsveir Journal.
• [Jun. 2023] Our paper “Robust Routing Made Easy: Reinforcing Networks Against Non-Benign Faults” was published by EEE/ACM Transactions on Networking.

Publications

Powered by Jekyll and Minimal Light theme.