In this session, we will discuss the concept of the Recurrent Inference Machine (RIM), a deep learning alternative to classical iterative inference algorithms wildly used in computational imaging. Not only does RIM implicitly learn a regularizing prior for the inverse problem, it also learns the iterative inference algorithm itself, leading to faster and more accurate results compared to Compressed Sensing and other Deep Learning approaches. Following an introduction of theory behind the method, we will be discussing applications to medical (MRI) and astronomical (strong lensing source reconstruction) imaging problems. Papers to be discussed:
- https://arxiv.org/abs/1706.04008
- https://www.sciencedirect.com/science/article/abs/pii/S1361841518306078
- https://arxiv.org/pdf/1901.01359.pdf
The Berkeley Statistics and Machine Learning Forum meets biweekly to discuss current applications across a wide variety of research domains and software methodologies. Hosted by UC Berkeley Physics Professor and BIDS Senior Fellow Uros Seljak, these active sessions bring together domain scientists, statisticians and computer scientists who are either developing state-of-the-art methods or are interested in applying these methods in their research. Practical questions about the meetings can be directed to BIDS Fellow Francois Lanusse. All interested members of the UC Berkeley and LBL communities are welcome and encouraged to attend. To receive email notifications about the meetings and upvote papers for discussion, please register here.
Speaker(s)
François Lanusse
While at UC Berkeley, François Lanusse was a Data Science Fellow at BIDS and a Postdoctoral Scholar with the Berkeley Center for Cosmological Physics (BCCP) and the Foundations of Data Analysis (FODA) Institute, exploring the intersection between cosmology, statistics, and machine learning. His research was focused on measuring and exploiting the gravitational lensing effect (in which distant galaxies appear distorted due to the presence of massive structures along the line of sight) with the development of novel tools and methodologies based on sparse signal representations, convex optimization, and deep learning. Before joining Berkeley, Dr. Lanusse worked as a postdoctoral researcher within the McWilliams Center for Cosmology at Carnegie Mellon University, after completing a PhD in astrophysics at CEA Saclay near Paris.