About me

Hi! I’m a PhD student at School of Computing, National University of Singapore, working on security and provable AI (my NUS page here). I am fortunate to work with Prateek Saxena and Kuldeep S. Meel as my PhD advisors.

My main research focus is trustworthy machine learning where I have worked on verification and testing approaches to quantitatively assess the reliability of machine learning models. I am interested in developing machine learning algorithms with provable guarantees (such as robustness, privacy and fairness), as well as improving the scalability and expresiveness of verification and testing algorithms for neural networks.

Trustworthy ML

[ICSE’21] PROVERO: proverò (I will prove it) and pro-vero (pro-truth)

• Following NPAQ, we develop a black-box sampling-based approach to quantitative verification where we focus on probabilistic guarnatees for robustness. We find that our tool can approximate the adversarial hardness, a measure based on how many adversarial samples exist for a given image that correlates well to how attacks perform. This work has been published at ICSE’21.

[CCS’19] NPAQ (Neural Property Approximate Quantifier)

• A framework with PAC-style guarantees for quantifying logically encoded properties over binarized neural networks. We demonstrate its use in three security-related applications: quantifying local robustness, success of trojan attacks and fairness. This work has been published at CCS’19.

---

Hosted on GitHub Pages using the Minimal theme