Combining neural modes of learning for handwritten digit recognition

An ADaptive Function Neural Network (ADFUNN) is combined with the on-line snapdrift
learning method in this paper to perform optical and pen-based recognition of handwritten digits.
Snap-Drift employs the complementary concepts of minimalist common feature learning (snap) and
vector quantization (drift towards the input patterns), and is a fast unsupervised method suitable for
real-time learning and non-stationary environments where new patterns are continually introduced.
The ADaptive FUction Neural Network (ADFUNN) is based on a linear piecewise neuron activation
function that is modified by a gradient descent supervised learning algorithm. It has previously been
applied to the Iris dataset, and a natural language phrase recognition problem, exhibiting impressive
generalisation classification ability without the hidden neurons that are usually required for linearly
inseparable data. The unsupervised single layer Snap-Drift is effective in extracting distinct features
from the complex cursive-letter datasets, and the supervised single layer ADFUNN is capable of
solving linearly inseparable problems rapidly. In combination within one network (SADFUNN), these
two methods are more powerful and yet simpler than MLPs (a standard neural network), at least on
this problem domain. The optical and pen-based handwritten digits data are from UCI machine
learning repository. The classifications are learned rapidly and produce higher generalisation results
than a MLP with standard learning methods.

Citation:
Kang, M. and Palmer-Brown, D. (2008) ‘Combining neural modes of learning for handwritten digit recognition’ Proceedings of Advances in Computing and Technology, (AC&T) The School of Computing and Technology 3rd Annual Conference, University of East London, pp.99-107.