EPSRC logo

Details of Grant 

EPSRC Reference: GR/J15087/01
Title: THE SUITABILITY OF THE ASSOCIATIVE STRING PROCESSOR FOR NEUROCOMPUTING
Principal Investigator: Glover, Dr R
Other Investigators:
Stonham, Professor T Stonham, Professor T
Researcher Co-Investigators:
Project Partners:
Department: Electronic & Computer Engineering
Organisation: Brunel University London
Scheme: Standard Research (Pre-FEC)
Starts: 01 October 1993 Ends: 31 December 1995 Value (£): 75,657
EPSRC Research Topic Classifications:
EPSRC Industrial Sector Classifications:
Related Grants:
Panel History:  
Summary on Grant Application Form
To investigate the implementation of neural networks using the ASTRA massively parallel associative processor developed by Aspex Microsystems Ltd. and Brunel University.Progress:ASTRA is essentially a Sun workstation host with a 16,384 processor single instruction multiple data stream associative processor closely coupled via an intermediate controller. Each processor has a 64 bit word of local content addressable memory. It has an interconnection network enabling word parallel one-to-many and bit serial many-to-many interprocessor communication.The lack of suitable systems software and programming development environment for ASTRA has hampered early progress for this particular application. However, the application software development is partially complete and important results are emerging.In semi-linear networks we have found that inner product calculation is the dominant factor in determining throughput and we are concentrating research on neural network structures where the novel, associative features of ASTRA may be exploited. Currently, Sigma-pi networks are under investigation. The Sigma-pi neuron model can be viewed as an associative element which is easily mapped on to the ASTRA associative string processor structure. The use of an associative processor allows the use of pre-calculated constrained look-up tables which makes training of the network more than thirteen times faster than for semi-linear networks. Results for 4-2-4, 8-3-8 and 10-4-10 networks show that between 1K and 13K million interconnections per second are achieved during training using a 20MHz clock speed machine. A data parallel mapping is currently under investigation that requires N copies of the net to be made in ASTRA to give an increase in learning speed of N.
Key Findings
This information can now be found on Gateway to Research (GtR) http://gtr.rcuk.ac.uk
Potential use in non-academic contexts
This information can now be found on Gateway to Research (GtR) http://gtr.rcuk.ac.uk
Impacts
Description This information can now be found on Gateway to Research (GtR) http://gtr.rcuk.ac.uk
Summary
Date Materialised
Sectors submitted by the Researcher
This information can now be found on Gateway to Research (GtR) http://gtr.rcuk.ac.uk
Project URL:  
Further Information:  
Organisation Website: http://www.brunel.ac.uk