disconnect.me.uk -- projects + code (!lowtek)

projects + research + code

Maps

Athens historic, neoclassical and eclectic architecture map
UK bus stop locations map

Weather

NeuroEvolution of Augmenting Topologies (NEAT)

NeuroEvolution of Augmenting Topologies (NEAT) is a neuroevolution technique -- a genetic algorithm for evolving artificial neural networks -- developed by Ken Stanley while at The University of Texas at Austin. It notably evolves both network weights and structure, attempting to balance between the fitness and diversity of evolved solutions. It is based on three key ideas: tracking genes with historical markings to allow crossover between different topologies, protecting innovation via speciation, and building topology incrementally from an initial, minimal structure ("complexifying"). -- wikipedia

java src -- NEAT applied to XOR problem

Betamax

betamax started off as a simple command line based 'chatter-bot' based on simple natural language engineering algorithms. Betamax was then connected to irc networks via pirbot java api, where it began to learn english. Using markov models to generate sentences from previously observed conversations, Betamax appeared to posses intelligence. Having had some fun, it was then decided to form a longer term project from Betamax - to explore machine conciousness.

Initially, using irc as an environment, betamax was improved by representing the meaning of sentences and observing the context of conversation. A java msn api was then developed based on protocol specifications from hypothetic.org to enable betamax to talk to individuals via msn. External tools used so far in the development of betamax include Stanford Log-linear Tagger for part-of-speech tagging and Wordnet - a "lexical reference system whose design is inspired by current psycholinguistic theories of human lexical memory"

A Heterogeneous Multi-Agent System For Urban Disaster Recovery

The RoboCup Rescue project was formed in 2001 following the success of the RoboCup soccer project. The project tackles the problem of real-world urban disasters, in which emergency forces (agents) must co-operate within a volatile and heavily constrained environment in order to minimise human casualties and structural damage. I've developed a Heterogeneous Multi-Agent System, using Neural Networks and Genetic Algorithms for tactical scheduling of resuce agents. It is designed to operate within the RoboCup Rescue urban disaster simulator.

java src

papers

Stubbings, P. et al, Modular Neural Networks for Recursive Collaborative Forecasting in the Service Chain, Journal of Knowledge Based Systems, (Elsevier Science, March 2008)

Stubbings, P., Shah, M., Collaborative Demand Forecasting in Service Chains, Book chapter in Service Chain Management, Technology Innovation for the Service Business. (C. Voudouris et al. Springer 2008)

E.P.K. Tsang, Cooperation in competitions -- constraint propagation strategies in chain-bargaining, Technical Report CSM-385, Department of Computer Science, (University of Essex, April 2003)

Stubbings, P., Machine Learning & Data Mining, A Comparative Study of The Backpropagation learning algorithm for Neural Networks & Decision Tree Induction, 2002