Open AccessMultimode Generative Adversarial Networks for Sequence Data Generation
Author(s) -
Weijie Kang,
Junjie Xue,
Jiyang Xiao,
Haizhen Zhu,
Jianfeng Li,
Changjun Li
Publication year - 2021
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1827/1/012209
Subject(s) - computer science , process (computing) , generative grammar , artificial intelligence , mode (computer interface) , range (aeronautics) , sample (material) , data mining , machine learning , algorithm , engineering , human–computer interaction , chemistry , chromatography , aerospace engineering , operating system
As a new type of artificial intelligence technology, generative adversarial network (GAN) has good data understanding and generation capabilities, and has a wide range of application prospects in the fields of image and speech. However, due to the lack of prior knowledge, its training process is less robust and prone to occur the pattern ignore. Its development is restricted to a certain extent, and its application scope still needs to be expanded. To solve the above problems, this paper introduces a knowledge confidence multimode GAN (KC-MGAN) algorithm, calculates the confidence of the input data through the reasoning method, and then puts the confidence and the input data into the GAN system to generate new sample data. During the training process, the confidence of the input data is continuously calculated, while the generated data samples are continuously evaluated. The training process will end until the GAN system reaches a stable condition. Finally, this paper takes the generation of UAV flight trajectory data as an example to verify the effectiveness of the proposed method. Some explorations have been made for the application of data generation and GAN’s training mode with the prior knowledge.