Computational Modelling Group

Associative learning in ecosystems: Network level adaptation as an emergent propery of local selection

Richard Watson, James Dyke (Investigators), Daniel Power

Ecosystems may exhibit collective adaptive properties that arise from natural selection operating on their component species. These properties include the ability of the ecosystem to return to specific configurations of species, in a manner highly analogous to mechanisms of associative learning in neural networks.

Centre for Doctoral Training in Next Generation Computational Modelling

Hans Fangohr, Ian Hawke, Peter Horak (Investigators), Susanne Ufermann Fangohr, Thorsten Wittemeier, Kieran Selvon, Alvaro Perez-Diaz, David Lusher, Ashley Setter, Emanuele Zappia, Hossam Ragheb, Ryan Pepper, Stephen Gow, Jan Kamenik, Paul Chambers, Robert Entwistle, Rory Brown, Joshua Greenhalgh, James Harrison, Jonathon Waters, Ioannis Begleris, Craig Rafter

The £10million Centre for Doctoral Training was launched in November 2013 and is jointly funded by EPSRC, the University of Southampton, and its partners.

The NGCM brings together world-class simulation modelling research activities from across the University of Southampton and hosts a 4-year doctoral training programme that is the first of its kind in the UK.

Challenging Topological Prejudice - Automated Airframe Layout Design

Andras Sobester (Investigator), Paul Chambers

Aircraft preliminary design scopes are drastically narrowed by topological prejudice. Modern aircraft have settled on the same 'tube plus wing and cruciform tail' type topology that has been adopted through their ancestry, with no scientific evidence that this layout is optimal. This research project poses the question:

“Given a topologically flexible aircraft geometry that is free of prejudice or bias, would a sophisticated multi-disciplinary optimization process yield a conventional layout?”

Differences between sexual genetic algorithms and a compositional cooperative co-evolutionary algorithm

Richard Watson (Investigator), Daniel Power

Different algorithms are representative of different genetic processes. This work explores how algorithms representing sexual recombination can solve certain problems that hill climbers cannot.

Do the adaptive dynamics of host-parasite systems catalyse or constrain sympatric speciation?

Richard Watson (Investigator), Daniel Power

Coevolutionary dynamics affect both parties evolutionary trajectories. When might these affect speciation? This project uses a simulation model to explore the issue.

Evolving Behaviour-Dependent Strategies in Agent Negotiations

Enrico Gerding, Markus Brede (Investigators), Darius Pepe Falahat

We use genetic algorithms to evolve trading strategies for iterative bilateral negotiations between buyers and sellers. In contrast to previous work we evolve purely reactive strategies that base decisions on memories of behaviour in previous negotiation rounds. A paper was written on this research and was published in the proceedings for the European Conference on Artificial Life 2013.

Generating Optimal Ensembles of Earth System Models

Simon Cox (Investigator), Elizabeth Hart, Andras Sobester

GENIE is an Earth system model of intermediate complexity. As with other climate models, the tuning of its parameters is essential for providing reliable long-term forecasts of Earth system behaviour. We apply a multi-objective optimization algorithm to the problem. The aim of the tuning exercise is to find the optimal values for the free parameters that produce and euqilibrium model end state with the closest fit to equivalent observational data.

Multi-objective design optimisation of coronary stents

Neil Bressloff, Georges Limbert (Investigators), Sanjay Pant

Stents are tubular type scaffolds that are deployed (using an inflatable balloon on a catheter), most commonly to recover the shape of narrowed (diseased) arterial segments. Despite the widespread clinical use of stents in cardiovascular intervention, the presence of such devices can cause adverse responses leading to fatality or to the need for further treatment. The most common unwanted responses of inflammation are in-stent restenosis and thrombosis. Such adverse biological responses in a stented artery are influenced by many factors, including the design of the stent. This project aims at using multi-objective optimisation techniques to find an optimum family of coronary stents which are more resistant to the processes of in-stent restenosis (IR) and stent thrombosis (ST).

Nonlinear Optics in Structured Material

Peter Horak, Neil Broderick (Investigators)

Structured materials such as photonic crystals, optical fibres, Bragg gratings etc. are the ideal material for nonlinear optics. Properly engineered materials allows one to control which nonlinear interactions are observed and enhanced whilst other nonlinear interactions can be neglected. This work looks both at fundamental ideas as well as the fabrication of devices for advanced telecommunications.

Self Interest & the Evolutionary Optimisation of Adaptive Trading Agents for Continuous Double Auctions

Frank McGroarty, Enrico Gerding (Investigators), Ash Booth

One cannot escape the recent crises in economics and the lack of understanding of financial markets that has been highlighted by them. Improvements to current market models are already being made and a realisation of the power of agent based modelling in such models is evident. In this project we seek to explore an existing model by Cliff of trader behaviour in continuous double auctions. We investigate the strategies that arise in such auctions when trader parameters are evolved with intent to maximise personal profit. Results show different trading strategies to those evolved by Cliff and explanations are given with regards to the self-interest.

Simulation of Parking Choice Behaviour

Ben Waterson, Hans Fangohr (Investigators), James Snowdon

Exploring how psychological models of individual parking search behaviours can be combined into an accurate simulation of vehicle flows, allowing for assessment of the impact on searching traffic of different demand/ supply ratios, different driver population characteristics and different charging regimes.

Spatially Embedded Complex Systems Engineering

Seth Bullock (Investigator)

SECSE brought together an interdisciplinary team of scientists working on an ambitious three-and-a-half year project titled. The research cluster spanned neuroscience, artificial intelligence, geography, and complex systems in an attempt to understand the role of spatial organization and spatial processes in complex networks within the domains of neural control, geo-information systems and distributed IT systems such as those implicated in air-traffic control.

The Origins of Communication Revisited

Jason Noble (Investigator), Jordi Arranz

Quinn (2001) sought to demonstrate that communication be- tween simulated agents could be evolved without pre-defined communication channels. Quinn’s work was exciting because it showed the potential for ALife models to look at the real origin of communication; however, the work has never been replicated. In order to test the generality of Quinn’s result we use a similar task but a completely different agent architecture. We find that qualitatively similar behaviours emerge, but it is not clear whether they are genuinely communicative. We extend Quinn’s work by adding perceptual noise and internal state to the agents in order to promote ritualization of the nascent signal. Results were inconclusive; philosophical implications are discussed.

The tarsal intersegmental reflex control system in the locust hind leg

David Simpson, Philip Newland (Investigators), Alicia Costalago Meruelo

Locomotion is vital for vertebrates and invertebrates to survive. Despite that feet are responsible for stability and agility in most animals, research on feet movements and their reflexes is scarce.
In this thesis, the tarsal reflex responses of locust will be studied and modelled with ANNs to achieve a deeper comprehension of how stability and agility is accomplished.
The choice of ANNs is linked to the applicability of the method into other fields, such as technological designs or medical treatment.