Neurosymbolic Deep Generative Models for Sequence Data with Relational Constraints

A key challenge facing deep generative models is the difficulty incorporating high-level structure into the generated examples--e.g., rhyming and meter across lines of a poem, or repetition across