Towards Large Language Models as Proposal Functions in a Neuro-Symbolic Expert System - 2022
Save Big on Coursera Plus. 7,000+ courses at $160 off. Limited Time Only! Grab it Explore a cutting-edge reasoning system that constructs structured text-based proofs of science facts grounded in an expert-verified external factbase. Delve into the NELLIE inference engine, which combines neural language modeling, guided generation, and semiparametric dense retrieval to replace handcrafted rules in a Prolog-style system. Discover how NELLIE dynamically instantiates interpretable inference rules to capture and score entailment decompositions over natural language statements. Gain insights into the system's motivation and search procedure, with a focus on how Transformer-based sequence models infuse semantics and structure from the factbase into the dynamic rule generation process, optimizing proof tree search. Based on research presented in a 2022 paper, this 51-minute talk from the Center for Language & Speech Processing at JHU offers a deep dive into the intersection of large language models and neuro-symbolic expert systems.