Fiche de WANG Qitong

Titre : Data Series Indexes that Learn
Date : 29 septembre 2023
Heure: 10:00
Résumé : Indexes are widely employed to speed up data series similarity search. Most indexes are based on summarized representations of the data series of lower dimensionality. In the study of this thesis, we found that both the summarization and the index structure might fail for some datasets. This is because their generation makes no use of the dataset and workload information. Hence, there is a pressing need for developing novel data series summarizations, and for designing data-adaptive indexing structure components (e.g., pruning filters). In this thesis, we explore how deep learning techniques can help data series similarity search techniques by make them more adaptive to dataset and workload. Both the summarization learning and index component (pruning filters) learning have been studied. We also successfully apply the proposed techniques in a real-world neuroscience problem. Future studies include to jointly learn the summarization and index structure.
Lieu : Room Turing, 7th floor Université Paris Cité - LIPADE, 45 rue des Saints-Pères, 75006 Paris