一維神經網路的局部擴散

Abstract

神經網路的局部擴散是由 Dr. Patrick Thiran 所提出，我們在一維神經網路最一般的形式去討論，我們想知道什麼條件成立時神經網路一定會有局部擴散，這期間我們引進了障礙概念，使得我們了解 Basic pattern 與局部擴散之間的關係，最後我們提出一個演算法來幫我們快速的判斷。

The local diffusion in CNN were introduced by Dr. Patrick Thiran﹒ We will discuss it in the general case of one-dimensional CNN﹒ We want to understand what conditions hold such that the CNN must have a local diffusion﹒ Here we introduce the block such that we get the relation between Basic patterns and the local diffusion﹒ At last we derive a algorithm to help us find the relation between basic patterns and the block﹒ And we prove that if the CNN has the block then the CNN has the local diffusion﹒