Plenary talks

Pier Luca Lanzi. Dipartimento di Elettronica e Informazione, Politecnico di Milano, Milano, Italy.

Learning to Play, Learning to Program, Learning to Learn (Experiences with Computational Intelligence for Simulated Car Racing)

Modern computer games are fun to watch and to play. Students love them! They are also difficult to program which makes them an excellent way to challenge students with difficult yet highly-rewarding problems. They pose great challenges to most methods of computational intelligence which makes them very good testbeds for research.

Car racing games are particularly attractive in that any car driver is a domain expert! In this talk, I will provide an overview of the recent research on the applications of computational intelligence to simulated car racing, including, development of drivers and driving behaviors through evolution, imitation and hand-coded design, evolution of tracks, and lessons learned from the recent scientific competitions.

Julian Francis Miller. Bio-Inspired Architectures Lab, Department of Electronics, University of York.

Evolving the brain inside the brain

Most of evolutionary history is the history of single cells. One of these cells is very special, it is called a neuron. Like other cells neurons are far from simple. In fact, a neuron is a miniature brain in itself. Neurons are not only very complex on the inside they also come in a vast range of complex morphologies.

Of course, natural evolution does not evolve brains directly. Instead it evolves genes. These genes represent complex 'programs' that cause the development of the entire organism (including the brain). All learning in the brain occurs during the development process.

So why do conventional Artificial Neural Networks (ANNs) represent neurons as extremely simple computational units in static networks? Why do they represent memory as synaptic weights?

Great advances in neuroscience have been made in recent decades and We argue that the time has come to create new models of neural networks in which the neuron is much more complex and dynamic. In such models, neural structures will grow and change in response to internal dynamics and environmental interactions. Like real brains, they should be able to learn across multiple domains without unlearning.

We review previous models and discuss in detail a recent new model and show that complex neural programs can be evolved that allow a developing 'brain' to learn in a number of problem domains.

Alan FT Winfield. Faculty of Environment and Technology. University of the West of England, Bristol.

Adaptive Swarm Foraging: a case study in self-organised cooperation

Inspired by the division of labour observed in ants, collective foraging has become a benchmark problem in swarm robotics. With careful design of the individual robot behaviours we can observe adaptive foraging for energy in which the swarm automatically changes the ratio of foraging to resting robots, in response to a change in the density of forage available in the environment, even though individual robots have no global knowledge of either the swarm or the environment. Swarm robotics thus provides us with both an interesting model of collective foraging, illuminating its processes and mechanisms, and a possible engineering solution to a broad range of real world applications, for example, in cleaning, harvesting, search and rescue, landmine clearance or planetary astrobiology. This talk will introduce the field of swarm robotics, using adaptive swarm foraging as a case study; the talk will address both the engineering challenges of design, mathematical modelling and optimisation, and the insights offered by this case study in self-organised cooperation.