Speaker: Christos Panourgias, PhD candidate
Title: Learning DTW-Preserving Embeddings for Scalable and Accurate Time Series Applications
Date: 2025-07-16
Time: 14:00- 15:00 (EEST)
Location: Orphanoudakis Seminar Room - FORTH, Main Building, 1st floor
Host: George Tzagkarakis, FORTH-ICS, SPL
Abstract:
Data series are central to numerous scientific and industrial applications, including medical signal analysis, financial forecasting, and sensor-based monitoring. A core challenge in working with such data lies in measuring similarity between sequences that may vary in length, exhibit temporal shifts, or contain local distortions. Traditional similarity measures like Dynamic Time Warping (DTW) provide effective alignment but incur high computational costs that limit their scalability. In response, research has increasingly focused on learning representations and distance functions that can approximate alignment-aware similarities while enabling efficient computation. Approaches leveraging deep neural encoders together with computationally efficient similarity measures have shown significant promise, allowing models to capture complex temporal dynamics in fixed-length embeddings and accelerate tasks such as retrieval, clustering, and classification.
Short Bio:
Christos Panourgias received his B.Sc. degree in Applied Mathematics from the University of Crete in 2022 and completed his M.Sc. studies in Artificial Intelligence and Computer Vision at the University of West Attica in 2024. He is currently pursuing a Ph.D. in the diNo group at Université Paris Cité where his research focus is on scalable techniques for data series analysis and representation learning with deep learning approaches.