Connectionists: (no subject)

[Stephen Welbourne]

Connectionists may be interested in the following PhD opportunity:-

Algorithms for large scale parallel distributed modelling of cognitive processes

Dr Stephen Welbourne & Professor Steve Furber

The Neuroscience Research Institute invites applications for a distinguished three-year PhD studentship commencing October 2008. The studentship is open to UK/EU nationals only due to the nature of the funding and will attract an annual stipend of £12, 940 in addition to tuition fees. 

Parallel distributed processing (PDP) models are widely used to model cognitive processes. These models are attractive to neuroscientists because they capture critical aspects of true neural processing, where computation is achieved through the interaction of a large number of simple processing units acting in parallel. Unfortunately, PDP models of cognitive systems are invariably modelled using serial hardware where the activations and weight updates for each component are computed sequentially. Consequently, as the networks become more complex the time required to train them increases, roughly in proportion to the square of the number of units. This places major limitations on the size of models that can be used, which in turn limits the types of cognitive behaviour that can be studied. However, through collaboration with the School of Computer Sciences we have a unique opportunity to tackle the problem directly, by utilising a novel parallel processing machine being developed as part of the EPSRC funded SpiNNaker project. 

The machine is designed for the modelling of spiking neural systems of up to a billion neurons. Consideration of the current PDP algorithms has shown that they too will map effectively onto the SpiNNaker hardware and its event driven model. In addition to the standard implementation of backpropagation, it will also be possible to use the system to explore more biologically realistic learning mechanisms such as hebbian learning.  Recent work suggests that multilayer generative models based on hebbian learning may allow for efficient learning even for very large networks with a deep structure. The SpiNNaker hardware will provide the perfect platform with which to explore this.

It is anticipated that the successful candidate will develop an extremely valuable combination of skills in parallel systems programming, computational neuroscience and machine learning. The project would therefore suit a numerate graduate with strong programming skills and an interest in neuroscience.

Applicants should hold a minimum 2:1 (or equivalent) degree in computer science, engineering or mathematics. Experience of using neural network models would be desirable.

Interested candidates should submit a CV, two references and detailed covering letter outlining their suitability for the project to Vicky Catterall, the NRI co-coordinator at 



Dr Stephen Welbourne
Neuroscience and Aphasia Research Unit (NARU)
School of Psychological Sciences (Zochonis Building) 
University of Manchester 
Brunswick Street
Manchester UK, M13 9PL

Tel 0161 3060442