Federated earth Learning

Cooperation Among Agents that Learn

"Communication and Learning in a Wide World"

This is what is going on

Current status of DistCBR and ColCBR is reported on Cooperative CBR.

Federated Learning: Motivation for a Framework for Agents that Learn

Problem solving in multi-agent frameworks offers new and challenging opportunities to learning approaches. We argue that learning in coordinated (agent-based) problem solving systems requires a more flexible architecture of learning than those usual in integrated multistrategy learning (IMSL) and we propose such a framework that we will christen federated learning.

The central difference between the federated learning approach and other approaches to IMSL is that while these other IMSL approaches have knowledge about when and how to learn hardwired into them, the federated learning approach has no such fixed knowledge. What a federated learning agent does know is how to find out about things it doesn't know. The result of this lack of "knowledge" (fixed decision about what to do) is greater flexibility in coordintaing with other agents and the ability to acquire new problem solving capabilities and improve performance.

The federated learning approach is based on and extends our previous work on learning as a introspective process in a reflective architecture (see Project ANALOG). We can summarize the federated learning approach as a natural broadening of the previous approach to deal with the issues of agenthood, namely:

From the perspective of problem solving systems, the role of learning is extending the range of problems amenable to be correctly solved and improving the systems performance.The role of federated learning in multiagent systems is improving problem solving of the collectivity of agents. This means we assume a cooperative behavior upon the agents in the system, and do not address the issues of learning in competitive multiagent systems.

[Plural] [Cooperative CBR] [Agents Page] [Features Terms] [Noos] [Team Members]