Snap-drift ADaptive FUnction Neural Network (SADFUNN) for Optical and Pen-Based Handwritten Digit Recognition

An ADaptive Function Neural Network (ADFUNN) is combined with the on-line snap-drift learning method in this paper to solve an Optical Recognition of Handwritten Digits problem and a Pen-Based Recognition of Handwritten Digits problem. SnapDrift [1] employs the complementary concepts of minimalist learning (snap) and drift (towards the input patterns) learning, and is a fast unsupervised method suitable for on-line learning and/or non-stationary environments where new patterns are continually introduced. The ADaptive FUction Neural Network (ADFUNN) presented in this paper [2, 3] is based on a linear piecewise neuron activation function that is modified by a novel 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 with no hidden neurons [2, 3]. The unsupervised single layer SnapDrift is effective in extracting distinct features from these 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, at least on this problem domain. We experiment with the Optical Recognition of Handwritten Digits and the Pen-Based Recognition of Handwritten Digits problems [4] from UCI repository. The problems are learned rapidly and higher generalisation results are achieved than a MLP.