About
Biography
I am Patrick Schäfer (Dr. rer. nat., Dipl. Inform.), a senior researcher at Humboldt-Universität zu Berlin at the Chair of Wissensmanagement in der Bioinformatik. My work centers on machine learning for time series data, scalable similarity search, and vector databases. I design algorithms with rigorous guarantees and turn them into reliable software artifacts such as the open-source aeon toolkit.
I obtained my doctorate at HU Berlin on scalable time series classification and have been part of research projects at Zuse Institute Berlin and in various industry collaborations focused on large-scale analytics.
Dissertation
Organizer roles
- AALTD Workshop — Advanced Analytics and Learning on Temporal Data (
'22, '23, '24, '25) - ECML 2024 tutorial: An Introduction to Machine Learning from Time Series
- Human Activity Segmentation Challenge @ ECML/PKDD 2023
- Guest editor for the Special Issue on Time Series Classification (2019)
Awards & nominations
- VLDB Distinguished Reviewer Award — 2025
- Two-time finalist for Humboldt-Universität zu Berlin’s “Preis für gute Lehre”: recognized in 2020 for Digital Teaching and in 2021 for Digitally Supported Teaching Concepts (finalists 2020, finalists 2021)
- Annual Scholarship from the German Academic Exchange Service (DAAD) — 2005–2006
- First place, ACM Programming Contest Berlin — 2005
Supervision
Bachelor theses: Time Series Motif Discovery; Visualizing Feature-Sets in TSC; Subword-Level TSC Framework; Analysis of TSC Classifiers with timeXplain; Ensembles for WEASEL/BOSS; SFA in MESSI; Motif Benchmark in Music/Lyrics; Open TSC with TEASER; Multivariate Segmentation with ClaSP; Motif Discovery in Audiobooks; Dilated Segmentation in ClaSP; Visual Features for Deepfakes; Multivariate Motif Discovery with Motiflets; Grouping Event Streams; Spikelet in ClaSP; Tennis-Stroke Classification with TSC Ensembles; Vulnerability Detection with LLMs; Deep Spatio-Temporal Land Cover Classification.
Study projects: Classification Score Profiles for Change Point Detection; Synthetic Time Series Generator for Motif Discovery; Latent Motif Discovery with Maximum Clique; Dilation in TSC; Open Set Recognition for Time Series.
Master theses: Synthetic Motif Generator; Time Series Segmentation; Variable-Length Latent Motif Discovery — Best Thesis Award 2022; Dilated Matrix Profile; Downsampling for TSC Visualization with ClaSP.
Research interests
- Time series algorithms: classification (BOSS, WEASEL), motif discovery (Motiflets, Leitmotifs), segmentation (ClaSP, ClaSS, ClaP), indexing & similarity search (SFA, SOFA)
- Vector search, high-dimensional indexing, and approximate nearest neighbor algorithms
- Experimental benchmarking and reproducible pipelines for ML systems
Service
- Program committees & reviews: KDD ’25 (Excellent Reviewer), VLDB ’25–’26 (Distinguished Reviewer ’25), ECML ’21–’24, AAAI ’21–’23, IJCAI ’20, ADBIS ’20, AALTD (since ’16), LDWA ’19, IEEE/CAA ’19, TKDE (since ’17), Data Mining and Knowledge Discovery (since ’16), IEEE Cybernetics (since ’18)
- Projects: Current — aeon (core developer). Past — sktime (former core developer), MoSGrid, Harness, Contrail, XtreemFS.
- Academic self-governance: Institute-wide space planning lead for Computer Science (SoSe 2022–present); Committee for Teaching and Study (Kommission für Lehre und Studium).