معرفی
Ari Karchmer is a Research Fellow at Harvard University, participating in the SAFR AI Lab led by Seth Neel and Salil Vadhan. He completed his PhD in Computer Science at Boston University under Ran Canetti, graduating in Spring 2024. His research focuses on theoretical computer science, machine learning, and cryptography, with particular emphasis on model extraction defenses, covert learning, and privacy-preserving techniques.
Education:
- PhD in Computer Science, Boston University (2024)
- BSc in Computer Science, Brandeis University (2018, Dean’s List)
Research Interests:
Ari's work bridges machine learning theory and cryptography. He explores foundational questions such as understanding why multimodal data benefits ML models, developing interpretable methods for data attribution, and analyzing computational limits of model extraction defenses. His research also addresses practical challenges like mechanistic interpretability and model editing.
Recent Article Trends:
His publications emphasize theoretical guarantees for observational defenses, computational separations in ML, and PAC-learning connections. Notable contributions include uncovering vulnerabilities in model extraction defenses and advancing component attribution techniques through linear models.
Awards:
- Best Student Paper Award at ITCS 2024
Advising & Grants:
Ari has no listed students but contributed to teaching fellowships at Boston University, including courses on Network Security, Algebraic Algorithms, and Responsible AI. His work has been supported by grants from the SAFR AI Lab and related initiatives.
Labs & Teams:
He is part of the SAFR AI Lab at Harvard, focusing on advancing AI safety and theoretical foundations. His collaborations include Seth Neel, Martin Pawelczyk, and Ran Canetti.
