卷积神经网络优化算法研究

Research on optimization algorithm of convolution neural network

为了解决卷积神经网络结构复杂,样本训练神经网络时间过长的问题,本文提出了采用分数阶理论优化卷积神经网络中的节点函数,使Sigmoid函数的收敛速度加快,在不影响卷积神经网络进行音频识别的正确率的前提下,减少了训练所需时间,达到提高整个神经网络的训练效率的目的。实验结果表明,在保证正确率的前提下该方法有效的减少了训练所花的时间,并可广泛应用于虚拟人运动控制系统中。

In order to solve the convolution neural network in speech recognition by samples to train the neural network takes a long time, this paper presents convolution neural network node function sigmoid function by using the theory of fractional order, accelerate the convergence speed of the sigmoid function, under the premise of the correct rate of speech recognition in does not affect the convolution neural network, so as to reduce the training time required to improve the training efficiency of the neural network. The experimental results show that the method can effectively reduce the time of the training when the guarantee is correct, and it can be widely used in virtual human motion control system.