COLLOQUIUM Princy Ranaivomanana, Radboud University & KU Leuven

Unsupervised machine learning for variability analysis in time-domain astronomy

The rapid progress in astronomical observation technologies has led to an enormous increase in light-curve data, bringing both exciting scientific opportunities and major challenges for time-domain astronomy. Traditional analysis techniques often fall short of fully exploiting the scientific potential of these large and increasingly complex datasets. As a result, machine learning algorithms have become essential tools for classifying, forecasting, and identifying patterns or anomalies in light curves. My research focuses on advancing our understanding of the formation and evolution of hot subluminous stars – hot, compact, and evolved stellar objects – by investigating their photometric variability using multi-band time-series data from multiple surveys and applying modern machine learning methods. In this talk, I will present statistical and machine learning approaches designed to efficiently handle unevenly sampled light curves for large stellar samples. I will demonstrate how unsupervised learning techniques can effectively detect and cluster different types of variability from irregularly sampled light-curve datasets. Applied to Gaia DR3 multi-epoch photometry, these methods reveal distinct groups, including hot subdwarfs, cataclysmic variables, and eclipsing binaries. These findings provide a strong foundation for future machine-learning-driven studies in time-domain astronomy and contribute to a deeper understanding of the formation and evolution of compact, evolved stars.

Talk in presence and via Zoom:

https://uni-hamburg.zoom.us/j/66006535328?pwd=aGkrSjJIYmZjK0VpYlpGL0ZrdHg2UT09