Compressed Machine Learning on Time Series Data

Detta är en Master-uppsats från Göteborgs universitet/Institutionen för data- och informationsteknik

Sammanfattning: The extent of time related data across many fields has led to substantial interestin the analysis of time series. This interest meets growing challenges to store andprocess data. While the data is collected at an exponential rate, advancements inprocessing units are slowing down. Therefore, active research is practiced to findmore efficient means of storing and processing data. This can be especially difficultfor time series due to their various shapes and scales.In this thesis, we present two variants for optimising a Greedy Clustering algorithmused for lossy time series compression. This study investigates, whether the efficientbut lossy compression sufficiently preserves the characteristics of the time seriesto allow time series prediction and anomaly detection. We suggest two variantsfor a performance optimization, Greedy SF and Greedy SAX. These algorithms arebased on novel lookup methods for cluster candidate selection based on statisticalfeatures of time series and extracted SAX substrings. Furthermore, we enabledthe clustering to allow processing time series with different value ranges, whichallows the compression of time series with various scales. To validate the endto-end pipeline including compression and prediction, a performance evaluation isapplied. To further analyse the applicability, a comprehensive benchmark against apipeline with an autoencoder for compression and a stacked LSTM for prediction isperformed.

  HÄR KAN DU HÄMTA UPPSATSEN I FULLTEXT. (följ länken till nästa sida)