Alternatively, you can request a copy by emailing me:
Thesis submitted as part of a taught course for the award of MSc in Artificial Intelligence at the University of Edinburgh.
Psychological and neurological evidence suggests that recognition memory might be composed of two processes; familiarity detection - categorisation of an item as familiar or novel, and recall - recognition of the stimulus. FamE, a neural network model which can carry out familiarity detection, was investigated and then implemented as a novelty filter for input from a robot mounted camera.
Investigation of the FamE model has (i) produced an alternative vector based explanation of how the FamE model works, (ii) indicated that only when large numbers of long binary input strings are processed will this model be more effective than alternative techniques and (iii) shown that the network's computational and memory requirements can be significantly reduced without effecting its performance.
The robot implementation demonstrated useful features of the FamE network in this application (i) it is able to reliably learning patterns from one presentation and (ii) it demonstrates a fair degree of invariance to noise and the position from which images are viewed.
Alternatively, you can request a copy by emailing me: |
|
