Neural Network Hyperparameter Tuning based on Improved Genetic Algorithm
Neural Network Hyperparameter Tuning based on Improved Genetic Algorithm
复制标题
基于改进遗传算法的神经网络超参数调优
DOI：
10.1145/3373509.3373554
复制
发表时间：
2019
期刊：
影响因子：
--
通讯作者：
Zhining You
中科院分区：
文献类型：
--
作者：
Xiang Wei;Zhining You
本文根据神经网络的结构特点，对传统遗传算法进行了一系列改进。该算法用于优化全连接神经网络中的一系列超参数，并寻找近全局最优的超参数组合。在使用MNIST数据集和20轮模型训练的情况下，使用该算法对全连接神经网络进行优化。实验表明，当种群进化到30代时，模型的准确率可达98.81%，高于官方样本模型的98.4%。这说明该算法在神经网络的超参数优化中起到了一定的作用。
In this paper, based on the structural characteristics of neural networks, a series of improvements have been made to traditional genetic algorithms. The algorithm is used to optimize a series of hyper-parameters in the fully connected neural network, and to find the near-global optimal combination of hyper-parameters. In the case of using MNIST data set and 20 rounds of model training, the algorithm is used to optimize the fully connected neural network. Experiments show that when the population evolves to 30 generations, the accuracy of the model can reach 98.81%, which is higher than 98.4% of the official sample model. This shows that the algorithm has played a certain role in the super-parameter optimization of the neural network.