Hans Farrell Soegeng

Hans Farrell Soegeng

Math PhD @ NTU

Singapore

Hi, I'm Hans! I am a second-year Mathematics PhD student in the SYLLAB group at Nanyang Technological University (NTU) Singapore, advised by Professor Thomas Peyrin. My research is supported by the Provost Graduate Award by the NTU School of Physical and Mathematical Sciences (SPMS). Previously, I earned my B.S. in Mathematics at NTU.

My research falls broadly at the intersection of interpretable ML and efficient ML. I focus on designing ML architectures that can be fully converted into human-understandable rules where the model's exact inference mechanism is reproducible, and minimizing the size of the rules or model to make the rules more understandable and inference cheaper. I like to work with simpler and smaller models that work and build fundamental things from first principles.

Beyond machine learning, I am interested in quantitative finance, specifically market microstructure and the role of market makers in providing liquidity across different asset classes, beyond traditional finance such as perpetual DEXs (HyperliquidHyperliquid) and prediction markets Polymarket Kalshi. This summer, I'll be joining OptiverOptiver as a Quantitative Researcher Intern.

I commercially deploy the ML models I developed at TT-logic.ai, a deep-tech startup founded by my advisor and backed by NTUitive (the VC arm of NTU). I previously interned at Micron in the Capital Planning team.

Feel free to reach out for collaborations or just a chat.

Research

TT-Sparse: Learning Sparse Rule Models with Differentiable Truth Tables

Hans Farrell Soegeng, Sarthak Ketanbhai Modi, Thomas Peyrin

ICML 2026

Sparse truth table-based ML architecture with discrete TopK differentiable relaxation convertible to low complexity Boolean rules.

Beyond Filter Pruning: Top-K Spatial Selection for Efficient Neural Networks

Sarthak Ketanbhai Modi, Hans Farrell Soegeng, Thomas Peyrin

ECCV 2026

Intra kernel structural pruning with TopK selection for low-resource vision inference.

Towards Global and Exact Interpretability for Few-Shot Tabular Learning via Generative Data Distillation from Foundation Models

Hans Farrell Soegeng, Tristan Guérand, Thomas Peyrin

Under Review

Statistical analysis of performance and exact complexity of foundation model distillation through synthetic generative networks.

Leveraging Foundation Models in Healthcare: A Distillation Approach to Interpretable Clinical Prediction

Hans Farrell Soegeng, Tristan Guérand, Thomas Peyrin

XAI4Science @ AAAI 2026

Interpretable classification through distillation of foundation models in data-scarce clinical setting.

Neural Network-Based Rule Models with Truth Tables

Adrien Benamira, Tristan Guérand, Thomas Peyrin, Hans Farrell Soegeng

ECAI 2023

A CNN-based ML architecture convertible to Boolean logic for interpretability and formal SAT verification.

Probabilistic Methods of Deterministic Theorems in Mathematics

Hans Farrell Soegeng, Wu Guohua

URECA @ NTU 2022

Ramsey's and van der Waerden's Theory on how orderly mathematical structures emerges from chaos in graph and arithmetic sequence colorings.

© 2026 Hans Farrell Soegeng

Last updated: June 2026