Abstract:
We investigate the impact of Differential Evolution Binary Particle Swarm Optimization (DEBPSO) on the configuration of learning models used to classify the shape prototypes generated by hierarchical neural-network models of vision. Visual cortex inspired models of vision build a dictionary of shape prototypes represented as a feature vector which is then used for classification of difficult feature invariant recognition problems. We show that high performance on invariant object-recognition tasks can be improved upon by configuring the learning models used in the classification of the shape prototypes utilizing DEBPSO. When regarding imprinted and random prototypes with the evolutionary algorithm configured learning models we show that a larger improvement can be achieved on the imprinted prototypes than the random prototypes. These results show a better way of classifying the shape prototypes used by the hierarchical models of vision and that imprinted prototypes do contain more useful information than random prototypes that was previously underutilized by the un-configured learning models used in prior research.
