Jiashu Tao (陶嘉澍)

Privacy-Preserving & Trustworthy ML Researcher

I am a Ph.D. candidate at the National University of Singapore advised by Assoc. Prof. Reza Shokri. My work focuses on better privacy quantification frameworks and techniques, as well as making machine learning models more reliable, transparent and aligned.

Email me Google Scholar LinkedIn
Privacy Membership Inference Trustworthy AI Interpretability Machine Learning
Portrait of Jiashu Tao

About

Research Focus

I formulate more realistic game-theoretic privacy definition, craft more useful data signals, and develop more accurate statistical tests to infer data membership from model outputs.

Academic Snapshot

  • Ph.D. Candidate, School of Computing, NUS
  • B.Comp. (Hons 1st Class) Computer Science, NUS
  • B.Sc. (Hons 1st Class) Applied Mathematics, NUS

Education

Doctor of Philosophy in Computer Science

National University of Singapore · Singapore · 2020 – 2026 (expected)

Advisor: Assoc. Prof. Reza Shokri

B.Comp. (Hons 1st Class) Computer Science & B.Sc. (Hons 1st Class) Applied Mathematics

National University of Singapore · Singapore · 2016 – 2020

GPA: 4.76 / 5.00 (CS) & 4.74 / 5.00 (Math)

  • Double Degree Program, Turing Program

Research & Professional Experience

SenseTime International — Algorithm Research Intern

Singapore · Dec 2019 – May 2020

QuantEdge Capital — Software Developer Intern

Singapore · Dec 2019

National University of Singapore Research Institute — Research Intern

Suzhou, China · May 2019 – Aug 2019

Bioinformatics Institute, A*STAR — Research Intern

Singapore · Feb 2018 – Nov 2018

Selected Publications

Full list on Google Scholar

Range Membership Inference Attacks

Jiashu Tao, Reza Shokri · IEEE Conference on Secure and Trustworthy Machine Learning (SaTML), 2025

Defines a new privacy notion and auditing framework that captures more comprehensive and realistic information leakage.

Machine Learning from Explanations

Jiashu Tao, Reza Shokri · Actionable Interpretability Workshop at International Conference on Machine Learning (ICML), 2025

Studies how expert explanations can be integrated into learning algorithms to improve model alignment with human intuition.

Directions of Technical Innovation for Regulatable AI Systems

Xudong Shen, Hannah Brown, Jiashu Tao, Martin Strobel, Yao Tong, Akshay Narayan, Harold Soh, Finale Doshi-Velez · Communications of the ACM 67 (11), 2024

Provides a roadmap for making AI systems auditable and compliant with evolving regulatory frameworks.

Towards Overcoming False Positives in Visual Relationship Detection

Daisheng Jin, Xiao Ma, Chongzhi Zhang, Yizhuo Zhou, Jiashu Tao, Mingyuan Zhang, Haiyu Zhao, et al. · British Machine Vision Conference (BMVC), 2020

Introduces perception algorithms that reduce false positives in large-scale visual relationship detection benchmarks.

Open-Source Software

Building accessible tooling so practitioners can audit privacy leakage before deployment.

Privacy Meter

GitHub forks badge GitHub stars badge

Most powerful Python library for auditing privacy risks in machine learning systems.

Diagram of the core membership inference engine capabilities in Privacy Meter

Teaching

Translating trustworthy ML research into engaging classroom experiences.

Teaching Assistant, CS5562 Trustworthy Machine Learning

National University of Singapore · Fall 2021, 2022, 2023

Course website

Get in touch

Office
COM3-02-18, Security Lab
NUS School of Computing
11 Research Link
Singapore 119391
jiashut@comp.nus.edu.sg Google Scholar LinkedIn NUS profile