R01 / Academic portfolio
Research
Two connected themes: governing AI through evidence about how systems reason and behave, and building trustworthy AI for scientific and human-centred domains.
Research / Two themes
Google Scholar
Two research themes.
Selected work is grouped by the contribution it makes, rather than by chronology.
-
R01
AI Governance
I develop technical foundations for governing AI through inspectable evidence: methods that make model behaviour interpretable and explainable, test whether reasoning is faithful, monitor what systems understand and rely on, and evaluate whether agents act for defensible reasons.
AI governance and audit Chain-of-Thought Oversight Should Not Treat Faithfulness as Monitorability 2026 / NeurIPS Normative governance NoRA: Evaluating Grounded Reasonableness in Visual First-person Normative Action Reasoning 2026 / NeurIPS Reasoning faithfulness Faithful-First Reasoning, Planning, and Acting for Multimodal LLMs 2026 / ACL Monitorability Evaluating LLM Understanding via Structured Tabular Decision Simulations 2025 / arXiv Explanation agreement EXAGREE: Mitigating Explanation Disagreement with Stakeholder-Aligned Models 2024 / arXiv Model multiplicity Practical and Efficient Rashomon Set Sampling for Model Interpretability 2026 / AISTATS Interpretability Exploring the Cloud of Feature Interaction Scores in a Rashomon Set 2024 / ICLR -
R02
Trustworthy AI for Science
I build trustworthy AI with domain experts, combining predictive performance with scientist- and stakeholder-centred evidence. Current work spans materials science, education, and healthcare.
Scientific explanations Diverse Explanations From Data-Driven and Domain-Driven Perspectives in the Physical Sciences 2025 / MLST Cheminformatics Regional Explanations and Diverse Molecular Representations in Cheminformatics: A Comparative Study 2025 / Intelligent Computing Materials discovery Multi-target Neural Network Predictions of MXenes as High-capacity Energy Storage Materials in a Rashomon Set 2023 / Cell Reports Physical Science AI in education Student-Centered AI Requires Auditable Learning Evidence Accepted / AACL-IJCNLP 2026AI tutoring Ownership Tracing in AI Tutoring: Correcting Apparent Mastery with Source-of-Practice Evidence Accepted / ICDM 2026 AI in education Learning-preserving AI tutors and responsible use of AI in higher education Active research direction Healthcare Trustworthy AI for clinical and neuroscience decision contexts Active collaboration
Academic portfolio