Publications

My research is driven by an interest in AI for science—using learning systems to model, understand, and reason about real-world phenomena. The papers below are grouped into three sections that each contribute to that goal: direct work on physical systems, approaches that emphasize structure and reasoning, and methods focused on robustness, feedback, and reliability. Together, they reflect how I think about building AI systems that are both scientifically grounded and practically useful.

AI for Physical Systems & Scientific Understanding

This work focuses on using machine learning to model, reconstruct, and reason about physical phenomena, particularly in fluid dynamics and flow-related systems.

Journal of Biomechanical Engineering, 2024

Unsupervised Denoising and Super-Resolution of Vascular Flow Data by Physics-Informed Machine Learning

Théophile Sautory, Shawn C Shadden

Uses physics-informed learning to reconstruct high-fidelity flow fields from noisy, low-resolution vascular data, emphasizing physical consistency and interpretability.

International Journal of Computational Fluid Dynamics, 2020

A Priori Sub-Grid Modelling Using Artificial Neural Networks

Alvaro Prat, Théophile Sautory, S. Navarro-Martinez

Explores the use of neural networks to model unresolved sub-grid phenomena in turbulent flows, integrating learning into physics-based simulation pipelines.

Structured Reasoning, First Principles & Compositional Systems

Work here focuses on structure, decomposition, and reasoning — bringing first‑principles thinking into broader AI settings.

International Conference on Logic Programming and Nonmonotonic Reasoning, 2022

Learning to Rank the Distinctiveness of Behaviour in Serial Offending

Mark Law, Théophile Sautory, Ludovico Mitchener, Kari Davies, Matthew Tonkin, Jessica Woodhams, Dalal Alrajeh

Applies structured comparison and ranking methods to behavioral data, emphasizing interpretability and principled reasoning over complex patterns.

AAAI Workshop on Hybrid Artificial Intelligence, 2021

HySTER: A Hybrid Spatio-Temporal Event Reasoner

Théophile Sautory, Nuri Cingillioglu, Alessandra Russo

Introduces a hybrid reasoning framework that combines learned representations with explicit temporal structure for event understanding.

Robustness, Feedback & Reliability in Learning Systems

This section looks at how learning systems behave under constraints, imperfect data, and feedback, which is central to applied AI, agents, and evaluation.

arXiv, 2024

Reinforcement Learning from LLM Feedback to Counteract Goal Misgeneralization

Houda Nait El Barj, Théophile Sautory

Investigates feedback-driven learning for large language model agents, focusing on alignment, failure modes, and robustness in goal-directed behavior.

SPIE – Automated Visual Inspection and Machine Vision IV, 2021

Data Augmentation and Pre-Trained Networks for Extremely Low Data Regimes in Unsupervised Visual Inspection

Pierre Gutierrez, Antoine Cordier, Thaïs Caldeira, Théophile Sautory

Studies robustness and generalization in low-data settings, combining pre-training and augmentation to improve reliability in real-world inspection tasks.

arXiv, 2021

Utilitarian Beliefs in Social Networks: Explaining the Emergence of Hatred

Houda Nait El Barj, Théophile Sautory

Models emergent behavior in social systems, analyzing how individual-level beliefs and interactions give rise to large-scale dynamics.