Google, IBM & Meta Certificates — All 10,000+ Courses at 40% Off
One annual plan covers every course and certificate on Coursera. 40% off for a limited time.
Get Full Access
Explore differentiable modeling techniques for machine learning in this plenary conference talk from JSALT 2025. Discover how automatic differentiation and backpropagation can be applied to differentiable computational graphs to incorporate scientific knowledge from mathematics, physics, and biology into deep learning architectures. Learn about current machine learning research themes including differentiable modeling, optimization of attention mechanisms in Transformer architectures, and large audio language models. Examine how making traditional forward models differentiable enables their integration into deep learning pipelines for parameter optimization, cost function minimization, inverse problem solutions, implicit neural representations, and explainable model development, particularly in data-sparse domains. See practical applications across computer graphics, human hearing, room acoustics, signal processing, and mathematical physics inverse problems through detailed example solutions and results presented by Professor Ramani Duraiswami from the University of Maryland, College Park.
