基于TensorFlow的深度神经网络优化方法研究

Research on Deep Neural Network Optimization Method Based on Tensorflow

深度神经网络属于机器学习领域的一项技术,实现了对高复杂性数据的建模.为了解决深度神经网络的过拟合问题,提高模型的鲁棒性,引入了正则化处理方法和指数加权移动平均算法,通过在损失函数中加入描述模型复杂化程度的因素,抑制模型在训练过程中可能出现的异常值,增强深度神经网络模型在未知数据上的健壮性.仿真实验结果显示优化方法有效可行.

Deep neural network is a technology in the field of machine learning,which realizes the modeling of high complexity data.In order to solve the overfitting problem of deep neural network and improve the robustness of the model,the regularization method and exponential weighted moving average algorithm are introduced.By adding the factors describing the complexity of the model in the loss function,the outliers that may occur in the training process of the model are suppressed,and the robustness of the deep neural network model on the unknown data is enhanced.Simulation and operation results show that the optimization method is effective and feasible.