AI Folk: Sharing Machine Learning Models in a Multi-Agent Community

Andrei Olaru, Alexandru Sorici, Mihai Nan, David-Traian Iancu

Abstract: With the rapid growth in the number of available pre-trained machine learning (ML) models for common tasks, with different performance, focus, and capabilities, complex problems can increasingly be solved through adequate choice of model, more than through training or tuning new models. In this paper we introduce the AI Folk methodology to address the challenge of autonomously managing ML models in a community of agents which can use and exchange semantic information about the models that they are using. We present a proof-of-concept implementation in an autonomous driving setting tackling various practical challenges which arise when dealing with this goal.