Exploiting Compositional Structure in Deep Learning - Theory and Applications

Watch a technical lecture from Johns Hopkins University professor Mauro Maggioni exploring both theoretical and applied aspects of deep learning at IPAM's Theory and Practice of Deep Learning Workshop. Explore a novel model for high-dimensional functions that extends beyond the single-index model by incorporating nonlinear projections onto one-dimensional curves, demonstrating how to overcome dimensionality challenges while maintaining optimal learning rates. Discover practical applications in cardiac digital twin technology, including a new risk prediction model combining clinical and imaging data, and an innovative learning architecture for predicting solutions to parametric PDEs across diffeomorphic domains with specific focus on heart electrophysiology predictions.