My research goal is to develop tools for building learning systems that generalise well out of distribution, do not fail spuriously and adapt to changing environments gracefully. I attempt to achieve this through the lens of representation learning using tools from generative modelling and optimisation. I've worked on algorithms to ensure robustness of deep learning models for example, being able to detect when such models are processing adversarial inputs, techniques for explainability of machine learning models, and causal inference for global health.