My overall goal is to design and build agents capable of knowledge discovery, particularly in the sciences. To this end, my current interests revolve around:

• How to achieve different types of machine reasoning: induction, deduction, causation, or decomposition. In particular, deep architectures, methods and systems for solving complex reasoning problems (e.g., mathematics, programming tasks, planning, etc.). My Master's thesis explored how language models can be used to solve multi-step reasoning problems via task decomposition as the complexity of the tasks increases.
• Generalizing reinforcement learning methods to better adapt to unseen tasks by meta-learning exploration and data-gathering strategies: "learning what data to learn from". Employing active learning frameworks to enable self-supervised agents to model the unknown dynamics of complex systems.
• Knowledge representations in deep models (e.g., latent embedding spaces) that enable reasoning, particularly using formal languages, hierarchical and object-centric compositional representation learning, etc.