Research
I work on machine learning under uncertainty and incomplete information, studied mostly through games. A full list of papers is on the publications page; this is just an overview of the main sub-areas.
Imperfect-information game AI
Many decisions are made without full information. I use games such as Reconnaissance Blind Chess to study decision-making in such environments.
- Neural Network-based Information Set Weighting for Playing Reconnaissance Blind Chess, IEEE ToG 2024
- Weighting Information Sets with Siamese Neural Networks in Reconnaissance Blind Chess, CoG 2023
- Supervised and Reinforcement Learning from Observations in Reconnaissance Blind Chess, CoG 2022
Preference and contrastive learning
Decisions can often be phrased as preferences. I develop contrastive and preference-ranking objectives that use this structure directly instead of forcing it into standard classification or regression.
- Contrastive Learning of Preferences with a Contextual InfoNCE Loss, 2024
- A Comparison of Contextual and Non-Contextual Preference Ranking for Set Addition Problems, ICML SubsetML 2021
- Predicting Human Card Selection in Magic: The Gathering with Contextual Preference Ranking, CoG 2021
Collectible card games
Magic: The Gathering is a hard machine learning domain: a constantly growing card pool, rules that shift with every set, making decisions with limited information. In addition, it is a great game. I cover learning card representations and LLM-based drafting.
- UrzaGPT: LoRA-Tuned Large Language Models for Card Selection in Collectible Card Games, 2025
- Drafting “Magic: The Gathering” With High-Dimensional Card Embeddings, IEEE ToG 2025
- Learning With Generalised Card Representations for “Magic: The Gathering”, CoG 2024
Reasoning models and neuro-symbolic work
Recurrent reasoning models, such as HRM and TRM, have recently made huge advances in combinatorial reasoning tasks. I research how to improve such models and how to integrate them into symbolic solvers.
- G-RRM: Guiding Symbolic Solvers with Recurrent Reasoning Models, NeSy 2026
- Symbol-Equivariant Recurrent Reasoning Models, ICML 2026