Collaborative Learning Through Shared Collective Knowledge and Local Expertise

Javad Mohammadi, Soheil Kolouri

Abstract: Collaborative lifelong learning for collective knowledge acquisition has recently attracted plethora of attention from various societies. Life long learning is rooted in continuous learning over a series of consecutive tasks. The learning tasks can be carried out by a single agent with access to a complete set of information or collaborative agents with restricted access to partial data sets. Existing lifelong learning methods are mostly based on centralized learning structure, which may limit applicability in most real settings. In this paper, we introduce a novel collaborative learning approach that relies on shared collective knowledge while preserving local expertise. In this collaborative lifelong knowledge acquisition each agent obtains agent-specific know-how, i.e., through learning from agent-specific tasks and has access to global knowledge accumulated by other agents dealing with similar tasks. The structure of our proposed solution ensures agreement among neighboring agents on a shared collective knowledge while maintaining a local expertise knowledge base. We tested our algorithm on several benchmark data sets.