Functional Diffusion

Functional diffusion models extend traditional diffusion models to generate data represented as continuous functions, rather than discrete points, enabling the creation of complex objects like images, videos, and 3D shapes. Current research focuses on developing efficient algorithms, often leveraging transformer or graph neural networks, to handle the infinite-dimensional nature of functional data and improve the realism of generated outputs, particularly in applications like drug discovery. This approach offers a versatile framework for diverse data types and irregular domains, promising advancements in generative modeling across various scientific fields and practical applications, such as AI-assisted drug design.