[1]陈 耀,宋晓宁,於东军.迭代化代价函数及超参数可变的生成对抗网络[J].南京理工大学学报(自然科学版),2019,43(01):35.[doi:10.14177/j.cnki.32-1397n.2019.43.01.005]
 Chen Yao,Song Xiaoning,Yu Dongjun.Iterative cost function and variable parameter generativeadversarial networks[J].Journal of Nanjing University of Science and Technology,2019,43(01):35.[doi:10.14177/j.cnki.32-1397n.2019.43.01.005]
点击复制

迭代化代价函数及超参数可变的生成对抗网络()
分享到:

《南京理工大学学报》(自然科学版)[ISSN:1005-9830/CN:32-1397/N]

卷:
43卷
期数:
2019年01期
页码:
35
栏目:
出版日期:
2019-02-28

文章信息/Info

Title:
Iterative cost function and variable parameter generativeadversarial networks
文章编号:
1005-9830(2019)01-0035-06
作者:
陈 耀1宋晓宁1於东军2
1.江南大学 物联网工程学院,江苏 无锡 214122; 2.南京理工大学 计算机科学与工程学院,江苏 南京 210094
Author(s):
Chen Yao1Song Xiaoning1Yu Dongjun2
1.School of IoT Engineering,Jiangnan University,Wuxi 214122,China; 2.School of ComputerScience and Engineering,Nanjing University of Science and Technology,Nanjing 210094,China
关键词:
生成对抗网络 迭代化代价函数 超参数可变 分布距离
Keywords:
generative adversarial networks iterative cost function method variable parameter distribution distance
分类号:
TP391
DOI:
10.14177/j.cnki.32-1397n.2019.43.01.005
摘要:
为了解决生成对抗网络(Generative adversarial networks,GAN)的训练难问题,该文在Wasserstein GAN(WGAN)方法基础上提出了迭代化代价函数及超参数可变的生成对抗网络。为了对原始WGAN中的惩罚项进行改进,用迭代的方法增加惩罚项代替原始随机选取的方法。针对WGAN中固定代价函数惩罚项的超参数,提出变动超参数策略,其变动的依据是仿分布和真实分布之间的距离。在MNIST手写字体数据集和CELEBA人脸数据集上的实验表明,与传统WGAN方法相比,该文方法在生成器的拟合速度上有了显著提高,充分验证了方法的有效性。
Abstract:
In order to solve the difficult training problem of generative adversarial networks,this paper proposes an iterative cost function and variable parameter generative adversarial networks based on the Wasserstein GAN(WGAN)method. For the improvement of penalty items in the original WGAN,iterative methods are used to increase penalty instead of the original randomly selected method. Aiming at the hyper-parameter of penalty item of fixed cost function in WGAN,the strategy of changing hyper-parameter is put forward. The change is based on the distance between imitation distribution and real distribution. Experiments conducted on MNIST handwritten font datasets and CELEBA face datasets show the effectiveness of the proposed method as compared with the traditional WGAN,significantly improving the convergence speed of the generator.

参考文献/References:

[1] 戚湧,胡俊,於东军.基于自组织映射与概率神经网络的增量式学习算法[J]. 南京理工大学学报,2013,37(1):1-6.
Qi Yong,Hu Jun,Yu Dongjun.Incremental learning algorithm based on self-organizing map and probabilistic neural network[J]. Journal of Nanjing University of Science and Technology,2013,37(1):1-6.
[2]Goodfellow I J,Pouget-Abadie J,Mirza M,et al. Generative adversarial networks[J]. Advances in Neural Information Processing Systems,2014,3:2672-2680.
[3]Goodfellow I. NIPS 2016 tutorial:Generative adversarial networks[J]. arXiv preprint arXiv:1701.00160,2016.
[4]唐朝辉,朱清新,洪朝群,et al. 基于自编码器及超图学习的多标签特征提取[J]. 自动化学报,2016,42(7):1014-1021.
Tang Chaohui,Zhu Qingxin,Hong Chaoqun,et al. Multi-label feature selection with autoencoders and hypergraph learning[J]. Acta Automatica Sinica,2016,42(7):1014-1021.
[5]Arjovsky M,Chintala S,Bottou L. Wasserstein gan[J]. arXiv preprint arXiv:1701.07875,2017.
[6]Arjovsky M,Bottou L. Towards principled methods for training generative adversarial networks[J]. arXiv preprint arXiv:1701.04862,2017.
[7]Gulrajani I,Ahmed F,Arjovsky M,et al. Improved training of wasserstein gans[C]//Advances in Neural Information Processing Systems. Long Beach Convention Center,PH:NIPS,2017:5767-5777.
[8]Radford A,Metz L,Chintala S. Unsupervised representation learning with deep convolutional generative adversarial networks[J]. arXiv preprint arXiv:1511.06434,2015.
[9]Zhao J,Mathieu M,LeCun Y. Energy-based generative adversarial network[J]. arXiv preprint arXiv:1609.03126,2016.

[10]Chen X,Duan Y,Houthooft R,et al. Infogan:Interpretable representation learning by information maximizing generative adversarial nets[C]//Advances in neural information processing systems. Center Convencions International Barcelona,Barcelona,Spain:NIPS,2016:2172-2180.
[11]丁绪星,朱日宏,李建欣.一种静止图像质量评价指标[J]. 南京理工大学学报,2004,28(5):507-510.
Ding Xuxing,Zhu Rihong,Li Jianxin. Image quality assessment index for still image[J]. Journal of Nanjing University of Science and Technology,2004,28(5):507-510.
[12]Zhang L,Zhang L,Mou X. RFSIM:A feature based image quality assessment metric using Riesz transforms[C]//2010 IEEE International Conference on Image Processing. Hong Kong,China:IEEE,2010:321-324.

备注/Memo

备注/Memo:
收稿日期:2017-12-06 修回日期:2018-05-16
基金项目:国家自然科学基金(61876072); 国家重点研发计划子课题(2017YFC1601800); 中国博士后科学基金特助(2018T110441); 江苏省自然科学基金(BK20161135); 江苏省“六大人才高峰”资助(XYDXX-012)
作者简介:陈耀(1990-),男,硕士生,主要研究方向:人工智能与模式识别,E-mail:yao_chen@aliyun.com; 通讯作者:宋晓宁(1975-),男,博士,副教授,主要研究方向:人工智能与模式识别,E-mail:x.song@jiangnan.edu.cn。
引文格式:陈耀,宋晓宁,於东军.迭代化代价函数及超参数可变的生成对抗网络[J]. 南京理工大学学报,2019,43(1):35-40.
投稿网址:http://zrxuebao.njust.edu.cn
更新日期/Last Update: 2019-02-28