Special Issue on High-performance Analysis of Nonstationary Scientific DataS
摘要截稿:
全文截稿: 2018-10-01
影响因子: 2.296
期刊难度:
CCF分类: B类
中科院JCR分区:
• 大类 : 计算机科学 - 3区
• 小类 : 计算机:理论方法 - 3区
Overview
Most systems existing in the real world have some degree of non-stationarity. For examples, physical and social phenomena in our daily life, such as climate change, public security, Earth surface processes, and brain activity are naturally time-varying events with frequency transients and complex harmonic interactions. Analysis of nonstationary scientific data has long been an important research challenge and hotspot in multiple disciplines.
The last decades have witnessed tremendous advances in signal processing to obtain reliable representations of the dynamics represented by nonstationary data. Time-frequency analysis methods rooted in linear and stationary models, such as the Fourier transform and Wavelet models, either ignore the data’s nonstationary nature or attempt to approximate the dynamics with the assumption of locally stationarity, which inevitably leads to bias in the conclusions. Chaos theory assumes that the data are captured from a chaotic system, but this is not always the case. It also heavily relies on prior knowledge and the empirical parameter settings. As possibly the most successful nonstationary method, empirical mode decomposition has difficulties in precisely separating desirable frequency features. Another open issue is handling the routinely intensive noises caused by the external forces of the observed system.
The recent advancements in data-driven analysis have provided an unprecedented opportunity to understand the dynamics and predict the behavior of nonstationary scientific data. Machine learning methods, especially deep learning, can adaptively explore the hidden features of the data without the need for presumptions on local stationarity or excessive prior knowledge about the underlying system. Models derived from representation learning can adapt well to different types and formations of data, which may be nonstationary in nature. Models receiving sufficient nonstationary training data can often achieve results of satisfactory precision. Empowered by the cutting-edge, high-performance computing technology and advanced data analysis method, it exhibits great potential for in-depth analysis of massive scientific data. Thus, there is a pressing need to present the state-of-the-art research on theories, algorithms, methods, and high-performance computing frameworks for nonstationary scientific models.
This proposal aims to collect eight articles (topics) coveringtheories, algorithms, methods, and computing frameworksfor nonstationary scientific models in connection with their applications in Earth sciences, meteorology, brain disorders, public security, special childhood education, and general data engineering problems. The theme of this proposal focuses on high-performance analysis of nonstationary scientific data. The potential authors will be solicited based on their publication records, collected from the Internet, from the past decade.
The topics of interest for this special issue include, but are not limited to:
- Basic theories for analysis of nonstationary data such as time-frequency analysis, Bayesian theory, deep learning, ensemble learning, MRF, Gaussian process, and etc
- High-performance spatiotemporal analysis of high-resolution remote sensing data
- Online multimodal social signal processing
- Surveillance video analysis in the smart city
- High-performance classification of multivariate electroencephalogram (EEG) data
- Massively parallel computing frameworks for high-dimensional time series
- Next-generation computing platform for high-throughput nonstationary scientific data