With the advance of sensor technologies, complex cyber-physical systems (CPSs) produce large amount of time series from spatial networks, which record the time-varying statuses of spatially distributed entities. Examples include traffic time series from road networks and chemical concentration time series from sewage pipe networks.
The Astra project aims at enabling accurate and scalable machine learning algorithms on such correlated time series, which improve the efficiency of the operations of the CPSs, e.g., reducing travel time in transportation systems and early warning of high concentration of toxic chemicals in sewage systems. In particular, the project invents learning algorithms to appropriately capture complex spatio-temporal dependencies among multiple time series and invents accurate and scalable analytics algorithms using the captured spatio-temporal dependencies. The accuracy, scalability, and relevance of the proposed algorithms are validated through industrial use cases.
Astra is a project at Center for Data-Intensive Systems (Daisy), Aalborg University. Astra is funded by the Independent Research Fund Denmark (IRFD) under agreement 8022-00246B and by the Department of Computer Science, Aalborg University.
Razvan-Gabriel Cirstea, Chenjuan Guo, and Bin Yang.
Graph Attention Recurrent Neural Networks for Correlated Time Series Forecasting.
5th Workshop on Mining and Learning from Time Series @ KDD'19 (MileTS19)
Tung Kieu, Bin Yang, Chenjuan Guo, and Christian S. Jensen.
Outlier Detection for Time Series with Recurrent Autoencoder Ensembles.
28th International Joint Conference on Artificial Intelligence (IJCAI2019)
Razvan-Gabriel Cirstea, Darius-Valer Micu, Gabriel-Marcel
Muresan, Chenjuan Guo, and Bin Yang.
Correlated Time Series Forecasting using Multi-Task Deep Neural Networks.
27th ACM International Conference on Information and Knowledge Management (CIKM2018)