en
Doctoral thesis
Open access
English

Information-theoretic frameworks for Machine Learning on Solar Data

Imprimatur date2024
Defense date2024
Abstract

This thesis is at the intersection of Information Theory (IT), solar physics, signal processing and Machine Learning (ML). It discusses the evolution of astronomy, highlighting the influence of technological advancements that, together with statistics, and the rise of Big Data in modern astronomy, enable efficient space exploration. It emphasizes the challenges and opportunities presented by ML in analyzing vast astronomical datasets, particularly focusing on solar data from NASA’s IRIS satellite. The thesis explores IT and ML applications for several detection problems, classification in decentralized systems, and forecasting multivariate data, such as solar activity. It addresses the challenges of astronomical data, with the design of sufficient statistics and information bottleneck formulations. The thesis emphasizes the importance of these formulations to enhance the scientific value learned on large collected datasets that are often multidimensional, sparse, weekly annotated, and processed with limited dedicated hardware as, for instance, in space-based instruments.

eng
Keywords
  • Information Bottleneck
  • Detection
  • Decentralized classification
  • Multivariate time series forecast
  • Solar data
  • Interface Region Imaging Spectrograph (IRIS)
  • Machine learning
  • Signal processing
  • Information theory
  • Rate-distortion theory
Funding
  • Swiss National Science Foundation (SNSF) - NRP75 Big Data [4075-167158]
Citation (ISO format)
ULLMANN, Denis Adrien. Information-theoretic frameworks for Machine Learning on Solar Data. 2024. doi: 10.13097/archive-ouverte/unige:177717
Main files (1)
Thesis
Secondary files (1)
Identifiers
50views
18downloads

Technical informations

Creation05/28/2024 2:50:37 PM
First validation06/11/2024 8:21:37 AM
Update time06/11/2024 8:21:37 AM
Status update06/11/2024 8:21:37 AM
Last indexation06/11/2024 8:21:55 AM
All rights reserved by Archive ouverte UNIGE and the University of GenevaunigeBlack