Learning to Count with Cell Assemblies: A Neuro-Symbolic Approach

Florin Leon

Abstract: Recent advancements in artificial intelligence have led to significant results in various domains, including image classification, natural language processing, and mastering complex games. However, current deep neural networks seem to process information differently from humans. Neuro-symbolic methods may offer a promising solution to address this concern. This paper proposes a preliminary cognitive architecture focused on neural cell assemblies, which can combine the adaptability of neural approaches with the explicit reasoning capabilities of symbolic systems. It presents a case study on learning to count, and highlights mechanisms for learning, generalization, and adaptation based on predictive errors.