Introduction to Unified Generative Modeling of 3D Molecules via Bayesian Flow Networks.
Abstract on the challenges in applying diffusion models to molecule geometry.
Proposal of Geometric Bayesian Flow Networks (GeoBFN) for 3D molecule geometry modeling.
Methodology on SE-(3) invariant density modeling and Bayesian Flow Networks.
Overcoming noise sensitivity in molecule geometry with smoother transformations.
Optimized discretised variable sampling for improved generation quality.
Results showing GeoBFN's superior performance in both unconditional and conditional molecule generation tasks.
Ablation studies on input modalities' impact on atom stability and molecule stability.
Unified Generative Modeling of 3D Molecules via Bayesian Flow Networks
"Advanced generative model derived from simplified continuity assumptions has been difficult to apply directly to geometry generation applications."
"GeoBFN achieves state-of-the-art performance on multiple 3D molecule generation benchmarks."