Research and Publications
Research Experiences & Interests:
My research interests have always been driven by following interesting problems that combine deep mathematical theory with real-world applications. My current research interests include:
📊 Machine Learning/Deep Learning/Statistics:
- LLMs (Interpretability, AI Safety, Agentic Systems, RL),
- Causal Inference/Causal Machine Learning/Causality, and
- Mathematical/High-dimensional/Applied Statistics.
Differential Geometry and Topology:
- Genericity & Transversality + Applications (to math/data science/machine learning).
- Riemannian & Neural Geometry + Applications (to mechanistic interpretability/AI safety/alignment).
These have evolved from earlier work as a graduate and an undergraduate student, where I have worked on projects involving:
- Hyperbolic PDEs and conservation laws,
- Mathematical general relativity,
- Bayesian modeling in nuclear astrophysics, and
- Numerical methods in quantum field theory.
Throughout, I think of myself as a “full stack mathematician” — I take projects end-to-end, from theory and modeling through implementation and simulation to production-scale systems whenever possible!
Publications:
* := Equal Contribution
† := First Author(s)
Underline := Myself
📜 Activation Flow: Manufacturing Activations for Steering
Hong Kiat Tan†, Linh Le, David Williams-King
Accepted at NeurIPS 2026 NeurReps Workshop & Submitted. Full version: arXiv:2609.32530 (2026).
Tags: Interpretability, Differential Geometry & Topology, ODEs, Numerical Analysis, AI Safety, LLMs.
📜 Generic Structural Stability for n x n Systems of Hyperbolic Conservation Laws
Hong Kiat Tan†, Andrea L. Bertozzi.
Submitted. arXiv:2609.28714 (2026).
Tags: Differential Geometry & Topology, Analysis of PDEs, Dynamical Systems.
📜 A Causal Model for Locating and Unlocking Sandbagging in Model Organisms
Hong Kiat Tan†, Linh Le, David Williams-King
Submitted. arXiv:2608.29461 (2026).
Tags: Interpretability, Causality, AI Safety, LLMs.
📜 Reference-Grafting Matches Fine-Tuning at Eliciting Sandbagged Capabilities
Linh Le†, Hong Kiat Tan, David Williams-King
Submitted. arXiv:2608.29458 (2026).
Tags: Interpretability, AI Safety, LLMs.
📜 Circuit Oracle: Automating Attribution Graph Analysis via Natural-Language Queries
Work done as a Research Fellow under SPAR.
Hong Kiat Tan*†, Shariar Kabir*†, Swastik Agrawal, Sai V R Chereddy, Sriram Balasubramanian.
Accepted at ICML 2026 Workshop on Compositional Learning: Safety, Interpretability, and Agents (Poster) and ICML 2026 Mechanistic Interpretability Workshop.
Tags: Interpretability, Multi-agent Systems, Causality, AI Safety, LLMs.
📜 Macroscopic Traffic Flow Network Modeling for Wildfire Evacuation: A Game-Theoretic Junction Optimization Approach with Application to the Lahaina Fire
Annie Lu*†, Hong Kiat Tan*†, Alex Xue*†, Alice Koniges, and Andrea L. Bertozzi.
Discrete and Continuous Dynamical Systems, 59 (2027), 123–184.
Tags: Optimization, PDEs on Graphs, Numerical Analysis, Mathematical Modeling.
🚦 GitHub Repo.
📜 Causal Lag Structure Discovery in Confounded Time Series via Orthogonalized Adaptive Estimation
Hong Kiat Tan†, Isaac-Neil Zanoria, James Chen, Haoyang Lyu, Mihai Cucuringu.
Submitted. arXiv:2610.05618 (2026).
Tags: Causal Machine Learning, Time Series Analysis.
📊 GitHub Repo.
📜 Beyond Statistical Changepoint Detection: Semantic Interpretation of Time Series via LLMs
Work done as an intern at Amazon,
Hong Kiat Tan†, Trilokya Akula, Akash Tonne, Thomas Blake.
ICLR 2026 Time Series in the Age of Large Models Workshop (Poster).
Tags: Mathematical Statistics, Multi-agent Systems, LLMs.
📜 Generic Structural Stability for 2 x 2 Systems of Hyperbolic Conservation Laws
Hong Kiat Tan†, Andrea L. Bertozzi.
SIAM J. Math. Analysis, 58(2) (2026).
Tags: Analysis of PDEs, Differential Geometry & Topology, Functional Analysis, Numerical Analysis, Fluid Dynamics.
📋 Slides.
📜 Weak Cosmic Censorship Conjecture for the Spherically Symmetric Einstein-Maxwell-Charged Scalar Field System
Hong Kiat Tan*†, Xinliang An*†.
Tags: Analysis of PDEs, Differential Geometry, General Relativity.
📜 Regularization of Complex Langevin Method
Zhenning Cai*†, Yang Kuang*†, Hong Kiat Tan*†.
Phys. Rev. D 105, 014508 (2022).
Tags: Numerical Analysis, Statistics, Quantum Mechanics.
📜 Hierarchical Bayesian Thermonuclear Rate for 7Be(n,p)7Li Big Bang Nucleosynthesis Reactions
Rafael S de Souza†, Hong Kiat Tan, Alain Coc, and Christian Iliadis.
The Astrophysical Journal 894, 132 (2020).
Tags: Bayesian Statistics, Astrophysics.
You can access my Google Scholar here.