Funded through EPSRC Research Studentships and British Telecom.
Supervisor: Professor Mike Holcombe, Verification and Testing Group.
[see my web pages for papers and details.]
This studentship has now been filled.
Francis studies social insects, including aspects of their organisation with emphasis on bees and ants. This is an area where there is some good theory (which matches the data) relating to how these "multi-agent" systems work e.g. in areas such as task partitioning. This CASE award would be initially about 50% social insect biology and 50% computer science, with the aim of drawing on this theory to better understand how multi-agent computational systems might be configured to perform tasks which exhibit desirable, adaptable behaviours. As the research progresses the emphasis may change.
The approach taken will be one whereby a number of different computational modelling techniques will be evaluated and the most appropriate one chosen to construct a more complete analysis of the opportunities for developing interesting and useful metaphors and algorithms of possible use to computing. Among the things that social insects do well is form reliable, robust systems that can easily be added to. These properties could well be useful in human systems and perhaps we can gain some insight from the insects in how to do this. One way to do this is to model a system and find rules that produce acceptable results over a wide range of conditions and with limited information about the exact nature of the current state of the whole system. In this respect, former PhD student here Carl Anderson showed that the regulation of numbers of foragers and receivers in the honey bee by making the appropriate dance (tremble dance to recruit receivers, waggle dance to recruit foragers) worked for a wide range of thresholds between making one dance versus the other.
Anderson CA 1998 Simulation of feedbacks and regulation of recruitment dancing in honey bees. Advances in complex Systems 1: 267-282.
The basic approach will be to consider each insect as an intelligent agent, whose behaviour is specified using some suitable notation. Candidates are: X-machines, L-systems and process algebras. The model will then need to be extended to allow for the modelling of a community of agents, communicating in clearly defined ways, so that the emergent behaviour from the community can be studied. It is important that these models reflect the real behaviour of insect communities observed in the laboratory. This should lead to a greater understanding of how complex communication systems might work in a robust and effective way. Insights from how ants communicate through the laying of pheromone trails have already demonstrated that useful algorithms for packet routing in networks can be found by looking at nature.
Some area of social insect life, such as the organization of foraging trails and pheromones, that can be investigated with computational methods and is relevant to both our biological and computing aims will be considered. The problem is explored in a way that is relevant to the biology, and hopefully can be novel computationally.
Some appropriate systems would be:
All three of these could generate, initially, simple situations to simulate and all are open to experimentation in the laboratory.
The biological problem will be generalized into a problem that can also be investigated using alternative computational techniques to see if any one technique is preferable. For example to gain insight into how we can generate a reliable system using limited-intelligence multi-agents etc. and good ways to model this. At the beginning of the PhD the student will get exposure to real social insects and work on one problem that is largely experimental, such as a study of ant foraging trails in order to get a feel for real ants and their foraging early on. Also at the beginning, the student gets exposure to the computing methods. After a year or two the student will put biological insights and computational insights, and the preliminary work above, together to generate and test novel biological predictions arising out of modelling and also the reverse--to use novel insights from social insects to build novel computational paradigms.