Human-centered AI

The gap is the whole discipline.

We build AI that reaches back on human terms — adaptive, interpretable, accountable for how it arrived at what it says.

Human-centered AI · two research programmes

Intelligence you can observe.

We build AI that models how people learn, decide and trust — adaptive systems that show their reasoning instead of asserting it.

Knowledge tracing
Per-learner mastery, estimated attempt by attempt
Interpretable tabular models
Bayesian networks and Markov blankets

Between human understanding and machine inference.

The MINN HAI Lab explores the synergy between the two — developing intelligent systems that are adaptive, interpretable and socially responsible, and that align with human values, cognitive processes and real-world needs.

Move the cursor — the lattice is the lab's Markov-blanket graph, not decoration
node in the blanket of the target variable
New research group · Human-centered AI

We model the learner, not the average.

Adaptive, interpretable, socially responsible systems — designed to perform and to be accountable for how they performed.

Enter the lab
The plane deforms around the cursor: a live surface, drawn from the model's own response curve.
2
research programmes
1
open learning resource, StatLab

Research

Two research programmes, and the publications behind them.

Programme 01

AI for personalised learning

Knowledge tracing: estimating what a learner has mastered, attempt by attempt, and reporting how uncertain that estimate is. The estimate is the model's claim about a person, so it is always shown with its band.

Publications
  • EIKTAIED, 2025
  • Privacy-Preserving Synthetic Data GenerationEC-TEL, 2018
  • IKTAAAI, 2022code
  • BKT-LSTMArxiv, 2021code
  • DSCMNPAKDD, 2019code
  • DKT-DSCICDM, 2018code
  • KTICDM, 2018
  • Q-matrix ReinmentEC-TEL, 2016
  • Uni Library Rec. Syst.e-Learning, 2013
Programme 02

Interpretable explanation of tabular models

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.

Publications
  • LAPLACEArxiv, 2023code
  • FSBN & SSBNICDM, 2023CXAI workshop
  • GBNCDSAA, 2014
  • GMBNCMLDM, 2014
  • MBNCComputer Science and Application, 2014in Chinese

Learning

Resources from the lab
StatLab

An interactive introduction to probability and statistics. Runs in the browser.

Open StatLab

Members

Principal investigator, researchers, students
portrait
TODO: name
Principal investigator
portrait
TODO: name
TODO: role
portrait
TODO: name
TODO: role
portrait
TODO: name
TODO: role
portrait
TODO: name
TODO: role

Funding

Sponsors and funding agencies

We would like to thank the following sponsors and funding agencies for supporting our research.

  • Uni Kyoto-Inria associate team grant — Inria, France
  • Natural Sciences and Engineering Research Council of Canada (NSERC) — Canada
  • National Natural Science Foundation of China (NSFC) — China
  • National Institute of Informatics visiting research grant — Japan
  • Singapore Management University visiting research grant — Singapore
  • Science and Technology Foundation of Xiamen — China
  • Science Foundation of Huaqiao University — China

Bring us a problem that needs an explanation.

sein.minn.cs @ gmail.com
MINN HAI LabHuman-centered AI