The NARNIAN Project: Communities of Agents Learning Over Time

NARNIAN is an innovative framework to study learning over time, fully inspired by Collectionless AI, where intelligent agents interact within dynamic environments and real-time data streams.

• Public announcement by Marco Gori and Stefano Melacci, March 21, 2025 (PDF)

• Take a look at an introductory presentation that explains the core concepts behind Collectionless AI and NARNIAN (PDF)

• The official code repository is now public (GITHUB)

What is NARNIAN?

Learning from data collections introduces risks related to data centralization, privacy, energy efficiency, limited customizability, and control}. Collectionless AI promotes on-the-edge local computations, where community of agents learn to interact in an efficient-and-controlled way with each other. This is the context in which the NARNIAN project is concieved, standing for NAtuRe iNspired Intelligent AgeNts.

NARNIAN is to Collectionless AI what training and benchmarking based on static datasets are to today’s AI. Actually, it is more general than that. NARNIAN is a dynamic conceptualization of the notion of learning setting and benchmark, though for studying Collectionless AI technologies but open and fully compatible with every other machine learning technology, allowing it to be be studied in a dynamic setting of continuous learning, interaction, progressive development.