Elastic Classifiers for Sequential Data

• Applicant: Professor Dr. Brijnesh Jain
• Subject Area: Image and Language Processing, Computer Graphics and Visualisation, Human Computer Interaction, Ubiquitous and Wearable Computing
• Term: from 2015 to 2021
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 290386697

Elastic learning methods in dynamic time warping (DTW) spaces are invariant under temporal dynamics (warping-invariant) and therefore often exhibit superior performance than their non-elastic counterparts in Euclidean spaces. Based on the results of the first funding-phase, this follow-up project aims at enhancing the elastic learning framework. The goal is to propose, analyze, and test elastic dimension reduction methods and elastic deep learning networks for time series classification. The main research question is: To which extent does warping-invariance affect the performance of a learning method? The purpose of elastic dimension reduction is to learn compressed representations of time series by filtering out noise. Introducing warping-invariance into deep learning is a way of incorporating prior knowledge from the time series domain into artificial neural networks. The proposed contributions continue to endow the mathematically rather poorly structured DTW spaces with novel warping-invariant enhancements of powerful learning methods from Euclidean spaces. We expect that the proposed approaches complement existing state-of-the-art techniques in time series classification.

• DFG Programme: Research Grants