Building Intelligent Agents with Neuro-Symbolic Concepts
A concept-centric framework for building agents that can learn continually and reason flexibly across multiple domains and offers several advantages, including data efficiency, compositional generalization, continual learning, and zero-shot transfer.