My PhD focuses on using concepts from Causality (more specifically, Actual Causality) to prove bounds and improve learnt representations (weights) of deep learning models, so that they learn concepts, rather than patterns, allowing for better generalisation of models in different contexts and domains. My work also aims to combine insights from continual learning, in order to address data and concept drift in models.
My work also spans the field of eXplainable AI (XAI), using a causal flavour to approach post-hoc explainability of models, this work has resulted in the development of the XAI tool, ReX, a post-hoc tool based on the concept of responsibility from Actual Causality.
The aim of my work is to build robust, interpretable systems that make safe, logical decisions, allowing for increased trust of automated systems in production.
I joined the STAI CDT, due to the alignment of my goals with mission statement of the CDT in building Safe AI systems. Moreover, the cohort-based approach of the CDT allows for individuals from diverse backgrounds to come together and collaborate, allowing for greater insights that can be used to build safer, more robust AI systems.
–
Undergraduate Qualification: BSc (Hons) Computer Science, American University of Sharjah
Masters Qualification: MSc Artificial Intelligence, King’s College London
Work Experience: Machine Learning Engineer (Intern), Kristal AI
Website: https://akchunyachanchal.com/