Disclosed are examples of systems, apparatus, devices, computer program products, and methods implementing aspects of machine learning for a decentralized content-centric system. In some implementations, predicted scores are obtained for a subset of fabric nodes of an overlay network. The scores can be based on prior training. A top scoring fabric node of the subset is identified as an egress node for communicating with a client. Also obtained are predicted scores for egress-origin segments defining paths in the overlay network between the egress node and origin nodes of the fabric nodes. The origin nodes store a content object part associated with digital content. One or more of the egress-origin segments is identified as being top scoring and is configured to be processed to identify one of the origin nodes for providing the content object part.