Data-Efficient Graph Grammar Learning for Molecular Generation

Explore a data-efficient neurosymbolic generative model for molecular generation in this 1-hour 8-minute workshop. Learn about a learnable graph grammar that generates molecules from a sequence of production rules, addressing the challenges of limited class-specific chemical datasets and the need to generate only physically synthesizable molecules. Discover how this method can be learned from significantly smaller datasets compared to common benchmarks, and how additional chemical knowledge can be incorporated through further grammar optimization. Gain insights into overcoming the limitations of deep neural network-based approaches that typically require large training datasets.