Auto Differentiation

Auto-differentiation is a powerful technique enabling the computation of gradients for complex systems, facilitating optimization in diverse fields. Current research focuses on extending auto-differentiation to previously intractable problems, such as Voronoi tessellations and relational computations, and improving its application within neural networks, including neural ODEs and spiking neural networks, often through novel loss functions and gradient estimation methods. This allows for more efficient training of complex models and enables data-driven solutions to challenging problems in areas ranging from materials science and drug discovery to robotics control and scientific content generation.