Hypothesis Transfer Learning for Distributed Knowledge Extraction by Intelligent Machines

In the context of industry 4.0, the data generatedby the production processes and collected by robots and intelligentmachines (RIMs) working nearby the production linesis a crucial asset if adequately exploited. Currently, such dataare collected and processed in powerful data-centres locatedin the Cloud to extract useful knowledge for improving theindustrial processes. However, due to potential technological andprivacy issues connected to exponential growth in the numberof connected devices, as well as to systems efficiency reasons,distributed knowledge extraction working closer to where datais generated are becoming a must. To this end, it becomesnecessary that RIMs collaborate to learn models in a distributed,collaborative and federated way. In this paper, we briefly describea solution of distributed learning based on Hypothesis TransferLearning suitable for such a context that allows RIMs to trainan accurate model collectively, also limiting the resources of thedevices involved in the process.