Research

We are working on the following research topics. Please feel free to contact us if you are interested in collaborating.

Theory of Computation

We study the mathematical foundations of computation through the lens of formal language theory. Our work focuses on the descriptional and computational complexity of finite automata and regular expressions, combinatorial properties of string similarity relations (in particular Simon’s congruence), and the inference of regular languages from examples.

Active Topics

Computing the edit-distance of formal languages
Descriptional complexity of finite-state automata and regular expressions
Regular language inference from positive/negative examples
Simon’s congruence: closure, neighborhood, and pattern matching on strings and languages
Closest and consensus substring problems for regular languages

Representative Publications

SplitRegex: Efficient Regex Synthesis by Neural Example Splitting, Journal of Automata, Languages and Combinatorics, 2025
Existential and Universal Width of Alternating Finite Automata, DCFS 2023; Information and Computation, 2025
Simon’s Congruence Pattern Matching, ISAAC 2022; Theoretical Computer Science, 2024
Automated Grading of Regular Expressions, ESOP 2023 (BK21 Top Conference, IF: 2)
On the Simon’s Congruence Neighborhood of Languages, DLT 2023
Closest Substring Problems for Regular Languages, Theoretical Computer Science, 2021
Reachability Problems in Low-dimensional Nondeterministic Polynomial Maps over Integers, Information and Computation, 2021

Formal Verification of Neural Networks

As neural networks are deployed in safety-critical settings, rigorous correctness guarantees become essential. We develop automata-theoretic and model-checking methods for formally verifying properties of deep neural networks (DNNs) and spiking neural networks (SNNs), with an emphasis on scalability, adversarial robustness, and efficient analysis.

Active Topics

Formal verification of deep neural networks and spiking neural networks
Model checking of neural network properties using automata-theoretic techniques
Efficient and scalable verification of neural networks using formal methods
Adversarial robustness analysis of spiking neural networks
STDP-based learning for adversarial robustness

Representative Publications

Timestep-Compressed Attack on Spiking Neural Networks through Timestep-Level Backpropagation, AAAI 2026 (BK21 Top Conference, IF: 4)

Programming Language Understanding and Generation

We investigate AI-driven approaches for understanding, generating, and repairing source code. We also study code complexity prediction using both structural program representations and deep learning, grammar-based test case generation, and the application of large language models to code analysis tasks.

Active Topics

Automated program synthesis and repair (NRF Basic Research Laboratory, 2023–2026)
Source code time complexity prediction using deep learning with automata-theoretic analysis
Automated test case generation from formal grammars and logical descriptions
Prompt engineering for code generation, analysis, and repair using large language models
Fault localization and automated code debugging

Representative Publications

LogiCase: Effective Test Case Generation from Logical Description in Competitive Programming, IJCAI 2025 (BK21 Top Conference, IF: 4)
CodeComplex: Dataset for Worst-Case Time Complexity Prediction, EMNLP 2025 (BK21 Top Conference, IF: 3)
EnCur: Curriculum-based in-context learning with structural encoding for code time complexity prediction, Expert Systems with Applications, 2026
MultiFix: Learning to Repair Multiple Errors by Optimal Alignment Learning, EMNLP 2021 (BK21 Top Conference, IF: 3)

Sports Data Analytics Using ML/AI

In close collaboration with industry partners, we develop data-driven methods for performance analysis in professional football (soccer). Our research spans multi-agent trajectory inference from sparse or heterogeneous data, contextual valuation of players and actions, formation and pressing analysis, and the use of reinforcement learning for strategic decision-making.

Active Topics

Contextual player performance evaluation using event and tracking data
Football formation and pressing analysis
Multi-agent trajectory imputation and prediction from event and snapshot data
Player and team valuation from action sequences using large language models
Multi-agent reinforcement learning for football strategy optimization
Transfer fit assessment and player scouting

Representative Publications

Valuing La Pausa: Quantifying Optimal Pass Timing Beyond Speed, MIT Sloan Sports Analytics Conference 2026 (Finalist — Top 7 of 200+ submissions)
Imputing Multi-Agent Trajectories from Event and Snapshot Data in Soccer, CIKM 2025 (BK21 Top Conference, IF: 3)
Trajectory Imputation in Multi-Agent Sports with Derivative-Accumulating Self-Ensemble, ECML PKDD 2025 (KIISE CS Top Conference)
exPress: Contextual Valuation of Individual Players Within Pressing Situations in Soccer, MIT Sloan Sports Analytics Conference 2025
Ball Trajectory Inference from Multi-Agent Sports Contexts Using Set Transformer and Hierarchical Bi-LSTM, KDD 2023 (BK21 Top Conference, IF: 4)
SoccerCPD: Formation and Role Change-Point Detection in Soccer Matches, KDD 2022 (BK21 Top Conference, IF: 4)