CUHK-Shenzhen · School of Data Science

Tsang Ka Wai

Associate Professor (Teaching) 曾家炜 · teaching statistics with AI

I teach statistics at CUHK-Shenzhen — and I am actively exploring how AI can help students learn statistics more clearly: not only how to compute, but how to explain what a model is allowed to claim.

Scroll
01 / Trajectory

Journey

My Trajectory

Hong Kong → Stanford → Academia Sinica → CUHK-Shenzhen. From mathematical training to a teaching-led career in statistics.

  1. 2004 – 2009

    The Chinese University of Hong Kong · Mathematics

    Three years of undergraduate study (BS, First Class Honours) and two years of M.Phil under the supervision of Prof. Jun Zou. This is where abstraction became a habit — and where I first learned that a clean proof and a clear explanation are different skills.

  2. 2009 – 2015

    Stanford University · Financial Mathematics & ICME

    MS in Financial Mathematics, then PhD in Computational and Mathematical Engineering under Prof. Tze Leung Lai. Stanford sharpened the link between stochastic models and decisions under uncertainty.

  3. 2015

    Academia Sinica · Postdoc · Visiting Instructor at Stanford

    Post-doctoral research with Prof. Ing Ching-Kang at the Institute of Statistical Science (Taipei, Taiwan), plus a short stretch as visiting instructor in Stanford Statistics — research and teaching side by side.

  4. 2015 – Present

    CUHK-Shenzhen · Associate Professor (Teaching), SDS

    Joined CUHK-Shenzhen in December 2015. I am now Associate Professor (Teaching) in the School of Data Science, coordinating the Statistics program and working on curriculum, advising, and how AI can support statistics learning.

Three things this line taught me

  • Use AI to enhance teaching and learning

    I am actively studying how AI can help students learn statistics — clarifying ideas, practicing judgment, and getting feedback — without replacing the teacher’s role.

  • Teach the decision, not only the formula

    Time series and forecasting courses work best when every method ends with “what claim can we make next week?”

  • Build the program, not only the lecture

    Curriculum committees, advising, and admissions are how a statistics education becomes a coherent path rather than a course list.

02 / Deep Dive

Teaching

Teaching that Leaves a Trace

I teach statistics so students can defend a claim — and I am exploring how AI can support that learning without replacing judgment.

CUHK-Shenzhen · School of Data Science

Associate Professor (Teaching) · Program Coordinator, Statistics

Dec 2015 – Present

Statistics teaching with a growing focus on AI for learning

Context

I regularly teach Time Series, Probability & Statistics, Forecasting and Predictive Analytics, and have also taught Survival Modelling, Statistical Capstone, and AI Exploration I.

Going further

Co-PI on AI-powered personalized learning (2025–2027) — connecting classroom experience with research on how AI can enhance teaching and learning in statistics.

Full teaching story →

03 / Work

Selected work

Research, Papers & Programs

Flip a card to see what each thread is really about — then open the full page.

Research

Post-selection Inference

Valid claims after the model has already been chosen from the data.

  • Time Series
  • Resampling
  • NSFC PI

Inference after selection

FocusPost-selection estimators for time-series models

GrantNSFC 12001461 · PI · 2021–2023

Why it mattersClassical intervals can lie once selection has happened; resampling keeps the claim honest

View details →

Publications

Papers that Travel

From Statistica Sinica to Statistics in Medicine — inference under dependence and selection.

  • Statistica Sinica
  • Technometrics
  • ORCID

Selected venues

SignatureHybrid resampling CIs · Statistica Sinica 2021

AlsoKnockoffs for streams · Adaptive enrichment designs

ORCID0000-0002-9383-4455

View details →

Collaboration

Personalized Learning

A teaching appointment meeting adaptive-learning research.

  • Co-PI
  • 2025–2027
  • Education × AI

AI-powered personalized learning

RoleCo-Principal Investigator

Period2025–2027 · 1+1+1 Joint Collaboration Fund

QuestionHow can personalization support statistical judgment instead of replacing it?

View details →
04 / Contact

Contact

Let’s talk

Teaching collaborations, advising questions, or research conversations on time series and inference.