带启发信息的蚁群神经网络训练算法  

h-ACO_R:An ACO_R Algorithm with Heuristic Information for Neural Network Training

作　　者：赵章明 冯径 施恩 舒晓村 

机构地区：[1]国防科技大学气象海洋学院,南京211101

出　　处：《计算机科学》2017年第11期284-288,296,共6页Computer Science

基　　金：国家自然科学基金(61371119)资助

摘　　要：ACO_R算法是一种应用于连续域实值优化的蚁群算法(Ant Colony Optimization,ACO)。ACO_R算法可用于训练神经网络。与常规的蚁群算法不同,ACO_R在训练神经网络时未考虑启发式信息(Heuristic Information)。在ACO_R的基础上,提出了一种将启发式信息与ACO_R相结合的神经网络训练算法——h-ACO_R。其中,启发式信息是通过计算神经网络的误差关于网络的权值向量的偏导数而得到的梯度向量。通过十折交叉验证方法,将h-ACO_R应用于UCI数据集中的zoo,iris和tic-tac-toe 3组数据的模式分类问题中来训练神经网络。与ACO_R相比,h-ACO_R算法在减小分类误差的同时能够提高收敛速度,其收敛的代数约为ACO_R算法的1/2,且经过完全训练,对zoo,iris和tic-tac-toe 3组数据的分类准确率分别为91.1%,93.3%和95.6%,高于ACO_R算法的83.1%,88.7%和91.9%。The ACO_R algorithm is an ant colony optimization（ACO）algorithm for real-valued optimization.The ACO_R can be used for training neural network.Unlike most of the conventional ACO algorithms,ACO_R does not consider heuristic information when training neural networks.So in this work,a new algorithm named h-ACO_R that incorporates the heuristic information into the framework of ACO_R was proposed for neural network training.The heuristic information in h-ACO_R is a gradient vector,which is obtained by computing the partial derivative of error term of the neural network with respect to weight vector.Using 10-fold cross-validation method,h-ACO_R is applied to train neural networks for pattern classification problems of zoo,iris and tic-tac-toe in UCI datasets.Compared with ACO_R ,h-ACO_R can reduce classification errors while speeding up the convergence process,with the average training generations of h-ACO_R being nearly 1/2 of that of ACO_R .After completely training by h-ACO_R ,the classification accuracy about zoo,iris and tic-tactoe are respectively 91.1%,93.3% and 95.6%,which have better performance than that of ACO_R ＇s 83.1%,88.7% and 91.9%.

关 键 词：蚁群算法 启发式信息 人工神经网络 神经网络训练 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]