AI for Human Behavior Modeling
People leave a record in the systems they use: what they attempted, what worked, what they came back to. We build models that read those records and describe the person behind them — what someone knows, and how that changes as they go. Most of this work is on learners, where the model's output is a claim about a person's understanding, so how uncertain it is matters as much as what it says.
- EIKTAIED, 2025
- IKTAAAI, 2022code
- BKT-LSTMArxiv, 2021code
- DSCMNPAKDD, 2019code
- DKT-DSCICDM, 2018code
- KTICDM, 2018
| Privacy-Preserving Synthetic Data Generation | EC-TEL, 2018 |
| Q-matrix Refinement | EC-TEL, 2016 |
| Recommender System for University Library | e-Learning, 2013 |
Causal Explanation for Human Understanding
Bayesian networks and Markov blankets: identifying the smallest set of variables that renders a target conditionally independent of everything else, so an explanation names variables rather than gesturing at importance scores.
Visual generative modeling
Learning what makes an image or video plausible well enough to generate new ones, not just classify or detect what's already there. A model that can generate a scene has learned something about its structure, so generation doubles as a test of whether a system understands what it sees.