3D ACNN; and its applications to maze problems

摘要

Artificial Cellular Neural Network (ACNN) is a kind of artificial neural network whose units are limited to connect just only their neighborhood units. When it comes to neural network, layered neural network and fully connected neural network are popular neural networks, but building them as IC is difficult since there are several problems; crossing of wires, delay of signals, and so on. On the other hand, a structure of ACNN can ignore these problems, so it will be easy to build large ACNN as IC. In this report, we demonstrate that ACNN has an ability to handle time sequential signals, through using "Sutton's maze" which is known as a difficult problem in the field of reinforcement learning.