Cognitive Science · computational modeling · machine learning
I am broadly interested in human cognition. Humans have remarkably limited cognitive resources, yet we can show creativity, reasoning, and problem-solving abilities that seem to go far beyond those limits. I hope to draw inspiration from the human mind to help build better AI systems.
Proceedings of the 48th Annual Meeting of the Cognitive Science Society (CogSci 2026). arXiv:2603.23244.
[CogSci 2026] [arXiv] [Task] [Data]
* Equal contribution.
arXiv:2605.09985.
[arXiv] [Experiment 1] [Experiment 2] [Data]
* Equal contribution. † Joint senior authors.
Human Factors in Communication of Design, AHFE Conference Proceedings. DOI: 10.54941/ahfe1006335.
Presented at the conference by Pinzhe Zhao.
I am designing a social-norm task where people create rules to coordinate robots on a grid. The project studies representational shifts: when people keep refining existing rules and when they reorganize them around a different feature.
I study how people build and reuse a library of helper patterns while solving visual puzzles. We compare human behaviour with computational models of library learning.
This project asks whether people form abstractions with future problems in mind. We use visual puzzle tasks and computational models to distinguish prospective learning from compression of past experience.
Visiting Student-RA at Shenzhen Medical Academy of Research and Translation. I worked on a multimodal machine-learning pipeline for predicting sinus pause events in mice from EEG, ECG, and sleep-state information. The project involved physiological signal preprocessing, feature extraction, sleep-state-aware analysis, and the training and evaluation of ensemble models for anticipatory risk prediction.
Voluntary Research Assistant at the Cognitive Ability & Plasticity Lab. I worked on machine-learning models for predicting individual differences in transfer learning from longitudinal cognitive training data. The analysis combined pre-training cognitive profiles, background variables, and task performance measures to model transfer effects, alongside diffusion-modeling parameters estimated with HDDM.