Collaborative Learning

The 4th industrial revolution adopts industrial Internet of Things (IoT) devices, i.e., robots, executing actions based on previous programming conditions. However, the learning systems do not exhibit complex tasks if the activities are not created with rules and hard-coded by humans or exhaustively trained with a large amount of labeled data. Therefore, while in Industry 4.0, robots reproduce actions previously programmed by software developers, machines and people will interact with each other in the fifth industrial revolution. The collaboration of things in autonomous systems will enable associating human interactions with cyberspace as a cooperative human-robot relationship. Thus, intelligent systems will discover answers to complex problems independently in the future.

In Collaborative Learning (CL), the problems can be spread in distributed systems and decomposed into minor problems scattered in multiple devices with enough computational resources. The dataset is split through different partners that train their separate models on each subset, share the learning, and update their parameters from a central aggregating service. This activity requires large bandwidths for communication and huge-energetic spending to support data movement in concurrent accesses. These constraints move IoT-

based solutions closer to Fog Computing (Fog) or Mobile Edge Computing (MEC), where storage and computation are near edge devices.

Our research focuses on Collaborative Learning related to AI algorithms, system design, and communication protocols for edge devices.