Shan He on Swarm Intelligence

The guest speaker yesterday was Dr Shan He, a lecturer in Computational Biology at the University of Birmingham. The subject of the talk was swarm intelligence, Dr Shan He’s area of research for several years. The talk demonstrated Dr Shan He’s research into simulating the ways animals aggregate into swarms, starlings for example. The presentation made good use of videos that really added to the explanations, despite early technical problems.

One aim Dr Shan He’s research was to understand how animals maintain swarm formations. Swarms have been modelled before, but Dr Shan He took a different approach, modelling using artificial neural networks. Using such a method requires far fewer definitions and depends strongly on the system learning by trial and error. The prey have a defined field of view, travel speed and that they should always flee from predators. After running numerous cycles the system is able to develop the most successful rules for survival and the different patterns types observed in nature can be seen in the computer model. After studying the resulting rules of the neural networks, it was found that many of the networks did not demonstrate either metric or topological models, but instead only interacting with the nearest two neighbours. When modelled using only the two neighbour method the same swarm patterns will still formed, but with far lower computational overheard than either metric or topological methods.

I have touched upon artificial neural networks during my studies of Artificial Intelligence and find the subject very intriguing. As a result I found this presentation very interesting and would like to look further into artificial neural networks. Programming without programming, allowing the system to take many different approaches to identify the best fit seems a little supernatural, but very effective. If the solution to a problem is highly difficult to obtain, but you have a significant volume of data to train an artificial neural network, then this provides a very beneficial solution. Allowing a system to learn by its mistakes and carry forward the most successful results allowed for replication of evolution that has occurred in nature. Surely this could also be used to predict other developments in nature; possibly the spread of disease or future evolution?

A very interesting presentation on artificial intelligence that due to the reference to swarms in nature was relatively easy to relate to.

Leave a Reply

Your email address will not be published. Required fields are marked *