My research focuses on developing advanced neuro-symbolic methods to identify and detect anomalies in the context of industrial processes. I plan to create a new formalism that can model all the relevant domain knowledge, taking into account historical data about past hazards. This representation will then be used to create a neuro-symbolic algorithm capable of classifying events as either normal or abnormal, automatically establishing a chain of causality in the latter case by virtue of the model’s symbolic nature. The resulting algorithm will finally be integrated into an early warning platform for use in green hydrogen processes.
In a world where large-language models are seemingly being used to solve problems across a wide range of domains, some sectors where safety is paramount — such as industrial processes and in particular hydrogen — are exempt due to their intrinsic high-risk nature. However we can exploit the trustworthiness and explainability of logic, combining it with the flexibility and speed of classical machine learning algorithms, to create a hybrid learning system that is safe, trusted and responsible when implemented properly. This goal is strongly in alignment with my interest in logic, and fits in perfectly with the STAI CDT’s goals.
–
Masters Qualification: MEng Computing, Imperial College London
Work Experience: Software Engineer at PerformanceStar in Santa Clara, California (2021-2024)