Multiagent Learning

Characterization of the Culture of Swarms

One of my goals over the winter break was to determine a good paper idea that had a lot of math and that would move the field of large multiagent learning systems forward.
Our contributions would be:

1. Alternative shapley value formula for swarms
2. Integration of the swarm value with k-order additive fuzzy measure theory
3. Solving the inverse problem to determine if a large system of agents is a swarm
4. Using the information to create autonomous hybrid swarm/multiagent learning environments

 

However, we might determine that the amount of work might be too much for one paper.

Culture

Could I use a model or description of the culture of an agent to determine how well it would adapt to working with others in a group.  Sort of a similarity index of cultures.  This would help an agent to decide if he wants to collaborate with a group.  Sean Luke has published two papers on how the use of culture has helped genetic programming (search culture http://www.cs.gmu.edu/~sean/papers/).

Skill Decomposition

In a cooperative multi-agent learning environment if a new agent enters the system the others can teach the new agent the high level task which it can then decompose to learn other useful skills.  In the single agent case an example would be teaching a robot to get from a to b which requires going through a door and the robot could learn the simpler activity of walking through a door by exploring its options space.  I think maybe task decomposition type algorithms could be augmented to decompose the option space.