研究では最急降下法に基づくＢＰ法を利用してプログラム開発を行った。
「非線形計画法」を利用するニューラルネットワーク学習において有効なアルゴリズムとして

・ＢＦＧＳ公式に基づく準ニュートン法
  Partial BFGS Update and Calculating Optimal Step-length 
  for Three-layer Neural Networks, Neural Computation (1997)

・ＲａｓＩＤ ( Ramdom Search with Intensification and Diversification )
  A New Random Search Method for Neural Network Learning - RasID - , 
  IEEE Trans. on Neural Networks (1998)

があり、これらを利用して改良していくことも可能ではないかと思う。
なお、ＧＡ（遺伝的アルゴリズム）を利用することはメモリと時間を浪費するので避けたい。

---------------------------------------------------------------------------------------------------

In my research, I have developed the programs 
by using the BP method (Back Propagation) based on steepest descent method. 
When you study the neural network learning by using "Nonlinear programming", 
some effective algorithms exist. 

・Quasi-Newton's method based on official BFGS 
  Partial BFGS Update and Calculating Optimal Step-length 
  for Three-layer Neural Networks, Neural Computation (1997) 

・RasID (Ramdom Search with Intensification and Diversification)
  A New Random Search Method for Neural Network Learning - RasID -,
  IEEE Trans. Neural Networks (1998) 

I think, we may be able to improve the programs by using these methods.  
However, because using GA (Genetic Algorithm) wastes the memory and time, I want to avoid it.