Date: June 29, 2022 

Source: College of Cyber Security 

A teacher and two students in JNU's College of Cyber Security published a research paper in the top journal Chaos, Solitons and Fractals in District 1 of the Chinese Academy of Sciences. The title is Chaotic signal denoising based on simplified convolutional denoising auto-encoder. The first author and second author are JNU master students Lou Shuting and Deng Jiarui respectively, with teacher Lyu Shanxiang as the corresponding author. The research results have been supported by Key Projects of National Natural Science Foundation of China, the Guangdong Provincial Basic and Applied Basic Research Major Project and other funding projects.

Chaos, fuzziness and neural networks are three major stimulated information processing systems. Studying nonlinear dynamics, chaos theory aims to discover the simple rules that may be hidden by seemly random phenomena, so as to reveal the common laws that a category of complex problems generally follow. Neural networks, brain waves and brain magnetic fields also carry chaotic characteristics. However, the observed chaotic signals are often contaminated by noise, which hides the system's real dynamic characteristics. 

For the first time, the authors considered chaotic signal denoising from the perspective of deep learning and proposed a chaotic signal denoising method referred to as simplified convolutional denoising auto-encoder. The method consists of a deep learning model with 13 layers, making it simpler than previous ones. Compared with conventional denoising techniques, the proposed method can achieve better signal-to-noise ratio, lower root mean square error and better proliferation exponents. The results show that the combination of chaos and deep learning has broad application prospects.

(Screenshots of the paper)

Link to paper: https://doi.org/10.1016/j.chaos.2022.112333