Manuel Lecha

Mathematical structure and abstraction are central to advancing the potential and applicability of modern artificial intelligence and machine learning systems. Integrating structured inductive biases, which reflect intrinsic data properties, has enabled deep learning models to achieve significant breakthroughs. Accurately capturing the complexity and dynamics of physical systems is critical in scientific domains, while robust reasoning fundamentally relies on structured representations and their underlying relationships. My research focuses on uncovering, understanding, and leveraging the essential structure inherent in these complex processes. By exploiting these mathematical insights, we aim to enhance model capacity and performance across diverse scientific and technological domains.