How do we design better drugs and materials with AI? Diffusion with Loop Guidance!

Unlocking novel generative capabilities with diffusion via “loop guidance”! In one of our  NeurIPS 2024 papers, we show how a novel framework “Diffusion Twigs”  (resembling multiple offshoots from a tree) opens extremely promising avenues for inverse molecular design and molecular optimization (fundamental tasks in de-novo drug discovery, material design, etc.). In contrast to the standard classifier-free and classifier-based guidance techniques, the proposed loop guidance approach is naturally tailored to hierarchical modeling, providing finer control over generation by learning to orchestrate the flow of information between processes for different properties.