Research output
Nanyu Luo highlighted · * equal contribution · # corresponding author. Use the toggle to reorganize by theme, venue, type, or year.
Journal of Educational and Behavioral Statistics · 2026
A hands-on framework for fitting Bayesian IRT models — dichotomous, polytomous, and multidimensional — directly in PyTorch and TensorFlow: accurate, low-bias, and a bridge between modern deep-learning tooling and psychometric modeling.
arXiv preprint · 2026
Robust uncertainty quantification for Bayesian analysis: the infinitesimal jackknife approximates bootstrap standard errors from a single MCMC run — no refits, no analytic derivatives — and, unlike the posterior SD, stays accurate under model misspecification across mediation, ANOVA, and multilevel models.
Journal of Educational and Behavioral Statistics (under revision) · 2025
FedIRT calibrates IRT models across institutions without ever pooling raw responses; a differentially-private extension (FedIRT-DP) adds auditable, student-level (ε, δ) privacy while matching centralized accuracy — released as an open-source R package.
Journal of Eating Disorders · 2025
A two-stage machine-learning pipeline (CNNs + LDA topic modeling) that detects and characterizes eating-disorder discourse in Chinese-language Weibo posts, surfacing five culturally specific themes to inform public-health surveillance.
Psychometrika · 2025
An adversarial variational-Bayes algorithm (AVB / IWAVB) for high-dimensional item factor analysis that lifts the restrictive Gaussian-latent assumption — scaling IRT estimation to complex, even multimodal, latent traits while matching or exceeding variational autoencoders.
EDM 2022 — International Conference on Educational Data Mining · 2022
Leveraging assignment-submission timing together with peer information to flag at-risk students early in online learning environments.