This project aims at developing new algorithms for fuzzy/flexible subsequence matching in time series. This can be achieved using new feature representations (e.g. Variational Autoencoders) and similarity/distance metrics. Furthermore, it is possible to use these representations to discover new patterns in the data using modern clustering techniques including Deep Learning, possibly resulting in new anomaly detection methods. Findings will be validated on clinical temporal data (ECG), as well as human activity data.
Improvement of fuzzy temporal pattern recognition, and anomaly detection, in clinical and human activity data (e.g. automatic segmentation of ECG signals or work cycles, respectively).
Author: Pedro Matias
Type: MSc thesis
Partner: FCT NOVA – Faculdade de Ciências e Tecnologia da Universidade Nova de Lisboa