Efficiency exploration of ALOPEX based recognition of hexagonalized images

The ALOPEX process, an optimization procedure which simultaneously varies several parameters based on single value feedback and noise application, is used to solve an image-recognition problem using several hexagonally based templates. Although in general the values of ALOPEX parameters are application-specific, the results show that the ALOPEX noise and feedback gain constants have optimum absolute sizes and relative ratios. In addition, the convergence improves by subjecting these parameters to an annealing process, provided the ratio of the constants are maintained near the ideal value. An attempt to enhance convergence using parallel processes with occasional crosstalk was also successful. Similar results were obtained with a different choice of the ALOPEX response function. It is suggested that these simple additions to traditional optimization methods such as ALOPEX can significantly improve the accuracy and time of convergence