Learning To Be Different

When building collective adaptive systems, such as sensor networks and swarms, it is normal practice to put the same software in each of the components that make up the collective. But in our recent research, we have shown that this can often be inefficient, leading to limited performance, and unnecessary resource overheads. Instead, it is often beneficial for there to be a diversity in the behaviours of the entities that make up the collective. But it is not good enough to just say that behaviour should be diverse: how should entities within a collective behave differently from each other? In our work, we have used online learning to generate effective diversity on an ongoing basis, over both space and time. We have shown that both these forms of diversity can be beneficial in a range of applications, including smart camera networks, and particle swarm optimisation, a search algorithm inspired by flocking birds. Using online learning to generate diversity helps the diversity stay adaptive, even in an environment full of change. This means that we can continue to maintain efficiency, and ultimately achieve better performance and lower power usage in a collective adaptive system, even as the world it inhabits changes.


Share this page: LinkedIn Twitter Facebook Google


Dr Peter Lewis is a lecturer and researcher in Computer Science at Aston University, Birmingham, UK. His primary research interest is concerned with adaptation, online learning and self-organisation in complex agent-based systems. He is a season ticket holder for Aston Villa FC.

ALICE: Aston Lab for Intelligent Collectives Engineering
EPICS: Engineering Proprioception in Computing Systems
Dr. Peter Lewis: Lecturer in Computer Science