Self-Supervised Learning from Noisy and Incomplete Data
About this Topic:
Many important problems in science and engineering involve inferring a signal from noisy and/or incomplete observations, where the observation process is known. Historically, this problem has been tackled using hand-crafted regularization (e.g., sparsity, total-variation) to obtain meaningful estimates. Recent data-driven methods often offer better solutions by directly learning a solver from examples of ground-truth signals and associated observations. However, in many real-world applications, obtaining ground-truth references for training is expensive or impossible. Self-supervised learning methods offer a promising alternative by learning a solver from measurement data alone, bypassing the need for ground-truth references.
In this webinar, the presenter will present a comprehensive summary of different self-supervised methods for inverse problems of recent years, with a special emphasis on their theoretical underpinnings, and presents practical applications in imaging inverse problems.
About the Presenter:
Julián Tachella (M’18) received the engineering degree from ITBA, Buenos Aires, Argentina, and the Ph.D. degree in electrical engineering from Heriot-Watt University, Edinburgh, U.K., in 2016 and 2020 respectively, with a thesis on image reconstruction methods for single-photon lidar devices.
He is currently a Research Scientist with the Centre National de la Recherche Scientifique (CNRS), based at the Laboratoire de Physique, ENS de Lyon, Lyon, France since 2021. From 2020 to 2021, he was a Postdoctoral Research Associate with the University of Edinburgh, working on deep learning approaches for computational imaging. He is also a Co-Founder of Blur Labs, a startup developing foundation AI models for computational imaging. In July 2023, he co-created DeepInverse (https://deepinv.org/), an open-source library for imaging inverse problems, for which he has served as a lead maintainer since. His research interests lie at the intersection of imaging inverse problems and machine learning, with a particular focus on self-supervised learning methods that can be trained without high-quality reference data, and foundation models.
Dr. Tachella is a member of the Technical Area Committee Theoretical and Methodological Trends in Signal Processing of EURASIP. He received the Best Paper Award at ICASSP in 2022, and his work was selected for oral presentation at AISTATS in 2024 and at CVPR in 2026.
Want to learn more about upcoming events & webinars?
Visit the events section of the Signal Processing website to see all upcoming lectures, workshops, webinars, and more.