Details of Grant 

EPSRC Reference:	GR/J44537/01
Title:	UNDERSTANDING AND USING NOISY SYNAPTIC WEIGHTS IN NEURAL LEARNING.
Principal Investigator:	Murray, Professor AF
Other Investigators:
Researcher Co-Investigators:
Project Partners:
Department:	Sch of Engineering
Organisation:	University of Edinburgh
Scheme:	Standard Research (Pre-FEC)
Starts:	06 June 1994	Ends:	05 October 1997	Value (£):	111,449
EPSRC Research Topic Classifications:
EPSRC Industrial Sector Classifications:
Related Grants:
Panel History:
Summary on Grant Application Form
1. To build a mathematical model of synaptic weight noise.2. To analyse various noise distributions.3. To study the measures of weight relevance, e.g. weight distributions.4. To determine the range of applications where noise will be beneficial. Progress:Incorporating weight noise into learning schemes is becoming a recognised technique for improving the performance of Multi-Layered Perceptrons. Based on our preparatory work, that underpins this project, several groups are using weight noise for learning enhancement in a variety of applications and neural architectures. Since the start of the present section of funded work in October 1994, we have progressed in our study of the underlying mechanisms of the noise by building a mathematical model. In addition, we have examined existing techniques for measuring the importance of individual weights to investigate how the noise influences the network storage mechanisms. In our work we have shown that the noise approximates a simple form of regulariser which gives a network solution that is smooth to weight perturbations. Previously we have shown that such a network property gives an increased tolerance to weight faults and in addition, by constraining the network solution, can give generalisation enhancements. Detailed analysis of our noise model is providing a clearer view of the mechanisms and will determine which of the models components are most influential. Therefore, we are progressing towards obtaining a full understanding of the noise that will allow us to judge how best to exploit its beneficial properties in any given application. At this early stage in the three year project we are already making significant inroads into fulfilling the objectives stated above.
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.ed.ac.uk