Joe Townsend

Department: Computer Science

Project Summary

My research is currently titled ‘Artificial Development of Neural-Symbolic Networks’.

Neural-symbolic integration concerns the representation of logic programs in neural networks. Researchers in the field tend to have one or a combination of the following aims: to produce powerful inference systems that are resistant to noise, to extract interpretable rules from information distributed across a trained network, and to work towards a working model of cognition. I am mainly interested in the third of these goals. The human mind is a neural structure that is capable of representing symbolic information, so Neural-Symbolic Integration is a logical step towards a cognitive model.

Artificial development is a form of evolutionary computing in which the genotype encodes rules for the gradual development of the phenotype (indirect encoding), rather than encoding an explicit description of the phenotype's structure (direct encoding). Indirect encoding is more biologically plausible than direct encoding, as DNA works in a similar way.

Since intelligence emerges through biological development, I believe that in order to produce a model of cognition, it is worth considering the possibility of producing such a model using a biologically plausible model of development. The aim of my research is therefore to produce neural-symbolic networks using artificial development.

Supervisory Team

Main Supervisor: Dr. Ed Keedwell

Secondary Supervisor: Dr. Antony Galton

Wider Research Interests

In addition to Neural-Symbolic Integration and Artificial Development, I am also interested in the more general field of machine learning. In particular, I am interested in neural networks and evolutionary computing.

Authored Publications/Reports

Joe Townsend, Ed Keedwell, Antony Galton (5th August 2013) Evolution of Connections in SHRUTI Networks, Proceedings of the 9th International Workshop on Neural-Symbolic Learning and Reasoning (NeSy13), 56-61

Joe Townsend, Ed Keedwell, Antony Galton (July 2013) Artificial development of connections in SHRUTI networks using a multi objective genetic algorithm, Proceedings of the fifteenth annual genetic and evolutionary computation conference (GECCO 2013)

Joe Townsend, Ed Keedwell, Antony Galton (13th April 2013) Artificial Development of Biologically Plausible Neural-Symbolic Networks, Cognitive Computation

Joe Townsend, Ed Keedwell, Antony Galton (July 2013) A Scalable Genome Representation for Neural-Symbolic Networks, Proceedings of the 1st Symposium on Nature Inspired Computation and Applications (NICA) at the AISB/IACAP World Congress 2012