I will introduce the All-atom Diffusion Transformer (ADiT), a unified generative modelling architecture capable of jointly modelling both periodic crystals and non-periodic molecular systems. ADiT is a latent diffusion model that embeds 3D atomic systems into a shared latent space, where it learns to sample new latents and map them to valid structures. ADiT achieves state-of-the-art performance for generative modelling across both molecules and materials, outperforming specialized system-specific methods while being significantly more scalable. I will show that scaling ADiT's model parameters predictably improves performance, towards the goal of a unified foundation model for molecular design.
Link to paper: https://www.arxiv.org/abs/2503.03965
Bio: "Chaitanya":https://www.chaitjo.com/ is a final year PhD student in Computer Science at the University of Cambridge, supervised by Prof. Pietro Liò. His research is about deep learning foundations for molecular modelling and design. He has previously interned at Prescient Design, Genentech and at FAIR Chemistry, Meta AI on the same.
