Greedy information acquisition in multi-layered networks

Abstract

In this paper, we extend our greedy information algorithm to multi-layered networks for improved feature detection. We have developed a new information theoretic network-growing model called greedy information acquisition. The method have shown good performance in extracting salient features in input patterns. However, because networks used in the method are single-layered ones, it has shown some difficulty in dealing with complex problems. In this context, we extend our greedy information acquisition method to multi-layered networks. By multi-layered networks, we can solve many complex problems that single-layered networks fail to do. The new algorithm was applied to two problems: the famous vertical-horizontal lines detection and a drive scene classification problem. In both cases, experimental results confirmed that our method could solve complex problems that single-layered networks fail to do. In addition, information maximization makes it possible to extract salient features in input patterns. The new algorithm can certainly contribute to the extension of neural computing.