|
In data intensive research applying correlation based algorithms, Bayesian approaches, and neuronal networks (e.g. deep learning) seem to be the gold standard of mathematical and statistical methods. In the proposed talk I would like to introduce a different way of gaining information from high frequency time series data by a symbolic and geometrical transformation. For sure, sequential data can be analyzed in multiple ways. My proposed approach is framing sets of data points and associating them to a contraction mapping, i.e. the procedure consists mainly of the following steps: a) Clustering (or boxing) data points of a temporarily frozen time period b) Associating a contraction map to each cluster c) Varying the clusters (which might lead to complex forms and complex dynamic). d) Each specific cluster and associated transition mapping parses a characteristic function which can be plotted and investigated.
The properties of the characteristic function can be used to investigate high frequency time series data. This will be demonstrated on a given high frequency data sample from cyber defense and/or systems reliability.
|
