Structural knowledge: from brain to artificial intelligence

Yingchao Yu, Yuping Yan, Yaochu Jin

Abstract: Organisms rely on structural knowledge derived from dynamic memory processes to adapt to the external world, employing cognitive maps and schemas. Similarly, artificial intelligence (AI) systems either explicitly or implicitly learn and utilize structural knowledge. Skillful utilization of this structural knowledge not only enhances the performance of AI models but also improves their transferability, generalization, and interpretability, crucial for developing robust and reliable AI systems. The importance of integrating insights from brain-based structural knowledge into AI systems cannot be overestimated. In this survey, we review research on structural knowledge in the brain, focusing on cognitive maps and schemas. We then examine computational models, highlighting potential mechanisms that serve as a bridge to AI. Finally, we provide an overview of AI models that leverage both brain-inspired and traditional structural knowledge and discuss how biological mechanisms can be applied to enhance AI systems. This survey aims to deepen the understanding of how structural knowledge in the brain can be constructed and utilized in AI, thereby bridging the gap between natural intelligence and artificial intelligence.

External IDs:dblp:journals/air/YuYJ25