Cong Ma
University of Chicago, IL·Mathematics & Statistics·Early-career
Saying "A is better than B" is often easier than assigning an exact score, which makes comparative, or preference, feedback a valuable resource for guiding recommendation systems, rankings and AI models, especially on subjective tasks where an ideal answer is hard to define. Real preference data, though, are noisy, sparse and heterogeneous: people answer inconsistently because of personal biases, skip items, or abandon a task out of fatigue.
- Fields
Mathematics & Statistics
Computer Science
- Topics
preference learning
preference feedback
pairwise comparisons
- Grants
- 1 known grant, through Sep 2030
What they work on
- Develop robust ranking algorithms for pairwise comparison data that tolerate ill-conditioned sampling of which pairs are compared and relax parametric modeling assumptions.
- Propose new estimation and uncertainty quantification methods for user-item rating systems that remain effective when observed ratings are sparse and users are heterogeneous.
- Introduce reinforcement learning from human feedback frameworks that incorporate "out-of-list" preference feedback and address distribution shift between training and deployment.
More about the research
This project develops robust statistical methods and theoretical foundations for learning from such data.
Why it matters
Recommendation systems, ranking tools and AI models trained on human feedback all rest on preference data that are incomplete and inconsistent. Methods that stay accurate under such conditions, and that report honest uncertainty, could make these decision-making systems more trustworthy across many domains, while clarifying the theory of when learning from comparisons succeeds.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Closest labs
Ravi Mangal
Colorado State University-Fort Collins, CO·Computer Science
Design quantitative metrics for evaluating code generated by large language models using symbolic execution, which explores program paths with symbolic inputs, and Bayesian program analyses that reason probabilistically about program behavior.
Surbhi Goel
University of Pennsylvania, PA·Computer Science·Early-career
Characterize why transformers learn brittle shortcuts rather than robust algorithms, using controlled mathematical environments that isolate real-world reasoning failures and remain tractable for rigorous analysis.
Morteza Dehghani
University of Southern California, CA·Computer Science
Develop video language models, AI that jointly interprets video and speech, to produce accurate and explainable summaries of officer-driver communication in body-worn camera footage of traffic stops.
The school
University of Chicago: 39 labs on PI Pages, $681M a year on research, 457 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.