«上一篇/Previous Article|本期目录/Table of Contents|下一篇/Next Article»

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:

30

期数:

2020年03期

页码:

7-12

栏目:

智能、算法、系统工程

出版日期:

2020-03-10

文章信息/Info

Title:

Study on Influence of Different Optimizers on Performance of LR under Gaussian Noise

文章编号:

1673-629X(2020)03-0007-06

作者:

徐龙飞; 郁进明

东华大学 信息科学与技术学院,上海 201620

Author(s):

XU Long-fei; YU Jin-ming

School of Information Science and Technology,Donghua University,Shanghai 201620,China

关键词:

机器学习; 线性回归; 优化器; 损失函数; 高斯噪声

Keywords:

machine learning; linear regression; optimizer; loss function; Gaussian noise

分类号:

TP301.6

DOI:

10. 3969 / j. issn. 1673-629X. 2020. 03. 002

摘要:

目前机器学习算法已经应用到社会的各个领域,如数据挖掘、信息个性化推荐和自然语言处理等,在人们的工作 和生活中起到了重要作用。 线性回归模型(LR)是常见的机器学习算法的一种,具有使用简单,容易理解,便于执行等特 点,但在加入噪声干扰的情况下模型性能会受到较大影响。 LR的优化方式包括批量梯度下降(BGD)、随机梯度下降 (SGD)、小批量梯度下降(MBGD)和Adam优化器等,最终训练后的模型性能受到优化方式、学习率、噪声等诸多因素的影 响。 为了研究在加入高斯噪声的情况下如何选择优化器来改善LR模型的性能,使用了Python语言和TensorFlow框架,通 过比较几种优化器的损失函数和计算时间来研究加入高斯噪声后对LR模型性能的影响。 实验结果表明,在加入高斯噪 声的情况下,使用Adam优化器得出的损失函数和计算时间优于其他优化器。

Abstract:

At present,machine learning algorithms have been applied to various fields of society,such as data mining,personalized information recommendation and natural language processing,which play an important role in people’s work and life. Linear regression model(LR) is one of several common machine learning algorithms. It is simple to use,easy to understand and easy to execute,but the performance of the model will be greatly affected when noise is added. The optimization methods of LR include batch gradient descent (BGD),stochastic gradient descent(SGD),mini-batch gradient descent (MBGD) and Adam optimizer,etc. The performance of the final trained model is affected by optimization methods, learning rate,noise and many other factors. In order to study how to select the optimizer to improve the performance of the LR model with Gaussian noise,we uses Python language and TensorFlow framework to study the impact of Gaussian noise on the performance of the LR model by comparing the loss functions of several optimizers and computing time. The experiment shows that the loss function and computing time obtained by Adam optimizer is superior to other optimizers when Gaussian noise is added.

常用功能

更新日期/Last Update: 2020-03-10