Pruning with interval arithmetic perceptron

Possibility and Necessity Pattern Classification using an Interval Arithmetic Perceptron

“…In the following, intervals and real numbers are represented by uppercase and lowercase variables, respectively [2,[7][8][9]]. An interval is defined as a pair of real numbers representing the range between an upper and a lower value, which are identified by superscripts U and L. One can define mathematical operations on interval numbers:…”

“…Iris and Breast Cancer tests have been studied both using all the original inputs and only two ones. In the second case, the two inputs, which have the property of maximising the recognition capability, have been identified using an algorithm described by Drago and Ridella [9]. Studying the problem with this reduced input number offers the advantage of supplying the reader with a visual easily-interpretable representation.…”

“…As regards figures, only inputs 3 and 4 have been considered; indeed, it has been shown that inputs 1 and 2 can be pruned while maintaining the ability to separate class 1 data with a perceptron [9]. Class 3 can be identified with a perceptron using only inputs 3 and 4, with the misclassification of three patterns; conversely, using all four inputs, only one pattern is misclassified.…”

“…Sixteen patterns featuring missing inputs have [9], it was shown that it is possible to prune all other inputs, with a classification error of 3.95%, which corresponds to 27 misclassified patterns, 17 of which are benign and 10 malignant.…”

Objective assessment of MPEG-video quality: a neural-network approach

“…The dimensionality of the input data space was further reduced with the feature-selection technique described in [20]. The eventual 4-dimensional feature space covered the quantities: p,(Nn-bits), ph(Xq-scale(l)), ph (Xmv(l)), and ph (Smv-dev-std).…”