Abstract
Time-series data are ordered sequences of real values and arguably constitute the most important seminar of data in the world. Virtually all the records in any industrial context have a time-stamp and naturally represent time-series measurements. The concrete applications involving time-series data are abundant, ranging from physiological sensors, astrological light intensities, up to financial and economical recordings.
The study of time series gave birth to a series of data mining challenges. To mention a few, searching is a task aiming at finding similar occurrences of a query pattern. In order to speed-up the searching time, researchers emphasized the need for time-series indexing methods [1]. Other challenging problems demand finding meaningful time-series similarity measures [2], as well as clustering the series. Yet, another historic problem with roots in econometrics necessitates the prediction of future values of a time-series [2]. One can also point out time-series classification, which refers to the process of learning from expert-labeled time-series data, in order to automatically predict the labels of future time series.
This seminar aims to empower students with absorbing and discussing state-of-the-art research in the fields of time-series mining. Each student will select a paper from a pool of recent publications, understand, present and learn to assess its weaknesses. In the end of the course, the students are expected to have acquired a basic knowledge on one of the main tasks involving time-series data.
[1] Esling, P., Agon, C.: Time-series data mining. ACM Comput. Surv. 45(1) (December 2012) 12:1-12:34
[2] Gooijer, J.G.D., Hyndman, R.J.: 25 years of time series forecasting. International Journal of Forecasting 22(3) (2006) 443 - 473