多层感知器的加速算法

The Speed Gradient Algorithm of Multi-Perceptron

基于多层感知器(Multi-Perceptron)的反向传播(BP:Backpropagation)算法,本文提出了一种新的改进算法——加速度算法。其基本思想是交替使用负梯度方向[S^t:=—△E(W^(?))]和加速梯度方向[S^k(?)=W^k—W^(k-2)(k≥2)]作为搜索方向。理论分析与模拟实验表明,与基本BP算法比较,它有很快的收敛速度。

Based on the Back Propagation Algorithm of Multi— perceptron, a new improved algorithm—the Speed Gradient Algorithm is proposed in this paper. The basic idea of this algorihm is that both negative — gradient directionS^k:=- ▽E(W^k) and speed gradient direction S^k:W^k = W(k-2)(k≥ 2) are alternately used to be the search direction. The theoretical analysis and the analogical experiments have been shown that compared with the basic back propagation Algorithm this new algorithm has faster weaken speed.