Time in Data Analysis
10 March 2023 - 12h15-14h00
Uni Mail (Room MR070) & Online (Zoom)
Registration mandatory - Under this link
Holistic Analysis of Trajectories in the social sciences
Matthias Studer, Geneva School of Social Sciences, Institut de démographie et socioéconomie
Many social sciences studies aims to understand or describe how trajectories unfold over time. These trajectories are often described by categorical variable coding the situations encountered by individuals. In this presentation, we discuss sequence analysis, a so-called holistic method, aiming to study trajectories in their entirety according to the timing of the states, their sequencing and the time spent in each situation. Technically, the method work by creating a typology of the trajectory. The presentation is illustrated through a study on school-to-work transition in Switzerland.
Data Processing Center in Geneva framework for variability classification in ESA Gaia mission
Krzysztof Nienartowicz, Faculty of Sciences, Department of Astronomy
The ESA Gaia mission is a multidimensional one: its main purpose is to create a three-dimensional map of our galaxy, including not only the most precise positions of sources of light but also their movement and changes in behaviour visible over the mission span (https://www.cosmos.esa.int/web/gaia/science). The biggest in space, gigapixel camera placed in L2 point at 1.5 million kilometers away from Earth is special but even more special is the data analysis that has to be conducted to convert data into time-series, then to analyze these billions of time-series to finally classify them and publish the biggest variable stars catalog in history. To be able to do so, Gaia Data Processing Center in Geneva, as part of Gaia consortium, developed and operates a dedicated system dealing with over ten billions of time-series for over two and half billion stars. I will present the philosophical, managerial, technical and scientific background of the approach we took in order to construct this one-of-a-kind platform for previous Data Releases and the steps needed to provide variable classification for the public Data Release in 2025.
Time Series in Medical Data Science : Challenges and Opportunities
Mina Bjelogrlic, Division of Medical Information Sciences (HUG), Human-Machine Interfaces in Clinical Settings (UNIGE Faculty of Medicine)
Medical data can be structured, unstructured, sparse, dense, or even missing. To value medical data in data-driven approaches strong semantics are needed. Time is at the center of this battle to grasp meaning, yet its representation is complex and challenging for many models including current state-of-the-art deep learning model architectures. Electrocardiograms (ECG) carry sequences of data recorded every milliseconds, whereas a patient history might have information recorded over decades, all depending on different time scales, dynamics, and dependencies. The talk will present current work done to build interpretable models for ECG automatic classification in collaboration with the cardiology department at the HUG.