Seminární skupina 02 předmětu Laboratoř elektronických a multimediálních aplikací

[Jakub Ryšavý]: Sequential Representations 30. 9. 2021

Abstract

Time series data are specific with their dependence on time (in comparison to other types of data). It may sound obvious; however, it leads to a huge problem, what is "the right" representation for them (to solve some learning task). Probably the right answer to the question is: It depends (on the task etc.).

It holds similarly for sequential data as it is the more general term. Or from the other side, the time-series data could be represented as sequence data (primarily not dependent on the time). The neighbor data points are dependent,
but the time could differ. In some cases, it may be better to represent time series without dependence on time.

One of the questions is how to choose a resolution without losing important information. It is similar to choosing a resolution in Computer Vision. We mostly process downscaled images (at least for computational complexity reasons). Time series data could be downscaled by the Symbolic Aggregate approXimation (SAX) method; however, we lose some potentially important information. Therefore, some improvements were introduced [1, 2, 3]. These techniques will be discussed during the main part of the presentation.

Seminář 18. 3. 2021 10:00

Readings

  1. New Time Series Data Representation ESAX for Financial Applications: https://www.researchgate.net/publication/4238133_New_Time_Series_Data_Representation_ESAX_for_Financial_Applications (it is not necessary to log in for download)
  2. Transitional SAX Representation for Knowledge Discovery for Time Series: https://www.mdpi.com/2076-3417/10/19/6980
  3. An Improvement of Symbolic Aggregate Approximation Distance Measure for Time Series: http://nugget.unisa.edu.au/jiuyong/An-Improvement-of-Symbolic-Aggregate-Approximation-Distance-Measure-for-Time-Series.pdf or https://www.semanticscholar.org/paper/An-improvement-of-symbolic-aggregate-approximation-Sun-Li/874ac1844aa0ed0f5f287bb2717a9459d7b713a6
Předchozí
Následující