Emergence of category-bias based learning learning acceleration in coherent neuralnetworks

摘要

By regarding the carrier frequency as an internal state in coherent neural networks, such a function is realizable by a self-organizing mechanism that identical input yields different outputs according to its internal state. In this paper, we analyze the learning acceleration by introducing contextual I/O patterns as a category bias condition. In the result, on the category bias condition including the contextual I/O patterns, there is a steeper learning acceleration than on the non-category bias condition. This result shows the potentiality that, based on this system, a self-organizing informaiton system that has human-like learning function will be realized.