Hey! I'm Amit Saha, a fourth-year undergrad at Georgia Tech studying computer science and math. I'm fortunate to work with Pan Li, Juba Ziani, Jacob Abernethy, and Salil Vadhan (via OpenDP).
My research focuses on the theoretical foundations of responsible learning. Specifically, I enjoy formalizing societal goals through {notions of trustworthiness}, understanding {statistical, computational} barriers which arise from these notions, and designing {accurate, efficient} algorithms under minimal assumptions. I seek to turn these insights into problem-agnostic principles for trustworthy algorithm design.
More broadly, I am interested in the complexity of learning and decision-making.
I will be applying to Ph.D. programs for admission in Fall 2027.
Conference Papers and Preprints
-
Adversarial Contextual Bandits via Private Online Confidence Oracles
Amit Saha, Jacob Abernethy, Pan Li, and Juba Ziani.
Under submission at ALT 2027. -
Privately Learning from Graphs: Privacy Amplification via Structured Subsampling
Amit Saha, Kevin Jacob, Pan Li, and Jan Schuchardt.
Under submission at AISTATS 2027. -
The Access-Similarity Lens: An Operational Copyright Framework for Generative Models
Amit Saha, Yinan Huang, Pan Li, and Eli Chien.
Accepted to TPDP 2026; under submission at SaTML 2027. * -
Evaluation Cards: An Interpretive Layer for AI Evaluation Reporting
Avijit Ghosh, Anka Reuel, …, Amit Saha, ..., et al.
NeurIPS 2026. -
Systematic Review of Recommended Practices for the Design, Conduct, and Reporting of AI Evaluations through 2025
Andrea Loehr, Ishan Khire, Yilin Huang, Drishti Sharma, Amit Saha, Yixiong Hao, and David Manheim.
EvalEval @ ACL 2026. -
Locality-Sensitive Hashing-Based Efficient Point Transformer for Charged Particle Reconstruction
Shitij Govil, Jack P. Rodgers, Yuan-Tang Chou, Siqi Miao, Amit Saha, Advaith Anand, Kilian Lieret, Gage DeZoort, Mia Liu, Javier Duarte, Pan Li, and Shih-Chieh Hsu.
NeurIPS 2025 Workshop on Machine Learning and the Physical Sciences.
* Earlier versions appeared at the International Conference on Machine Learning (ICML), MemFM + R2FM Workshops, June 2025.
Current Work
Recently, I've been thinking about (1) private in-context learning, (2) information-computation gaps in private statistical estimation under privacy and sparsity, and (3) computational rates for query-efficient private and replicable learning.
I've also been learning about the Sum-of-Squares hierarchy and its applications to the design of algorithms for statistical inference.
Beyond Research
Outside of my main research interests, I am deeply invested in {AI safety, autoformalization for mathematics via language models}, although I do not actively work on research in these fields. To this end, I help organize the AI Safety Initiative at Georgia Tech, and regularly make use of autoformalization tools for my own (mathematical) research.
Academics: I credit much of my research to ideas first learnt in different classes at Georgia Tech. If you're interested, check out my coursework here.
Industry Experience: I spent two wonderful summers working at Adobe on the Camera Raw team as a machine learning research intern under Michelle Qi and David Franzen.
Everything else: At last — nonacademically — I enjoy making pizza, reading, and listening to music all the time.
Contact
Feel free to reach out to me regarding {research, collaboration, advice} via e-mail at asaha92@gatech.edu.