I studied for a DPhil in Evolutionary and Adaptive Systems in the CCNR at the University of Sussex. My supervisors were Ezequiel Di Paolo and Phil Husbands.
I did a 4 year Maths degree (MMath) at the University of Nottingham, moving on to an MSc in Evolutionary and Adaptive Systems in COGS at the University of Sussex in Brighton, UK before continuing there with my DPhil.
I am interested in evolving agents which have certain desirable properties observed in animals, especially adaptability to and stability under changing conditions within and without the agent.
My DPhil project involved the design of two variations on Continuous-Time Recurrent Neural Networks which allowed for changes in the structure of the network over time ("growth" during the lifetime of the behaving agent). I tested these networks on various tasks involving sensor-motor co-ordination and maze solving. I found that the networks were difficult, but not impossible to evolve to perform simple tasks. Some of my results suggested that it is possible that controllers such as these may be more robust to certain types of disruptions in their morphologies.
In my earlier research I studied the role of homeostasis in producing unusually stable agents, and used large neural networks (1000+ neurons) with a view to developing tools for understanding complex systems in natural and artificial contexts.
I started a project to build a supercomputer out of normal PCs on the Sussex canvas, which I called The Sussex openMosix Cluster Project, but which never grew out of being a prototype.
I may be contacted on andybalaaam at artificialworlds.net (substitute the @ symbol for `at', and remove the spaces).
Balaam, A.J. (2006) "Exploring Developmental Dynamics in Evolved Neural Network Controllers" DPhil Thesis [ps.gz] [pdf]
Balaam, A.J. (2003) "Developmental Neural Networks for Agents" in Proceedings of the 7th European Conference on Artificial Life 2003 [ps.gz] [pdf]
Balaam, A.J. (2001) "nBrains: A New Type of Robot Brain" in Proceedings of the 6th European Conference on Artificial Life 2001 Springer-Verlag Heidelberg
The source code made available here is to facilitate understanding and replication of the work described in the publications above. You may view, modify and use this code under the condition that it is treated in the same way as a piece of academic work: i.e. you should give me credit if you use my work.
The code is developed under RedHat Linux 9, compiled with GCC 3.2.2 and uses wxWindows for GUI stuff and MySQL for data storage. The parallel genetic algorithm library PGAL, developed by Ian Macinnes (and me, somewhat) is used. It might be possible to convert PGAL to work on Windows, and my code should migrate fairly easily, but I haven't tried. Other Un*x versions should be fine.
Included are many different configuration files to run different experiments in the 'cfg' directory. The script 'runExperiment.sh' can be modified to use the config file you want to use. Configuration options may also be passed to the Evolver program directly. To view evolved agents use the Agentviewer program. Before you run any experiments you need to run 'pgal_db_setup' and 'agentviewer/createLifetimeDB.sh'.
Feel free to email me with questions about anything, including making this code compile and work. I'll do my best to help.
To reproduce the results of the 2003 paper (hopefully!) use the 2 configuration files 'cfg-NNDevtSOS10CP-Discrimination.txt' (for the developmental networks) and 'cfg-DRNN52-Discrimination.txt' (for the CTRNNs).
My code: nndevt.tar.gz
PGAL (the version I'm using): pgal.tar.gz
Here are some links somewhat or completely irrelevant to my DPhil:
| FreeGuide | My open source TV guide program. |
| Andy Balaam | My home page. |