21. November 2018

14.30 o'clock,
ARI Seminar Room, Wohllebengasse 12-14, Groundfloor

Automatic Transformation of Empirical Data Distribution as an Experimental Pre-Processing Step for Neural Networks - Pavol Harar

Abstract: Pavol Harar - University of Technology, Brno

In this talk we introduce the Redistributor. A direct algorithm for automatic transformation of data from arbitrary empirical source distribution into chosen target distribution. The method is based on approximation of source cumulative distribution function and is suitable also for machine learning scenarios. Transformation is continuous, piecewise smooth, monotonic and invertible. We demonstrate its usage on mel-spectrogram data used to train a neural network.