Electrical Engineering Trailblazers Seminar

Abstract: Computational imaging systems combine optical hardware with reconstruction algorithms to push the limits of image capture and interpretation. In these systems, measurements are designed for computational processing rather than direct human interpretation. As a result, the information available in measurements matters more than their visual interpretability.

We introduce a data-driven framework for imaging system design that quantifies the information content of measurements. Our mutual information estimator evaluates and optimizes imaging systems using only measurements and noise characterization, without requiring reconstruction, system modeling, or ground truth data. We show that these information estimates reliably predict system performance across imaging modalities, including radio astronomy and lensless imaging. We then establish design principles and optimize phase masks for lensless imaging, revealing that optimal designs are object-dependent and should be tailored to object sparsity. We conclude by outlining extensions to microscopy and black-box optimization. This talk illustrates how information theory provides powerful and practical tools for imaging system design.

Bio: Leyla Kabuli is a final-year PhD Candidate in Electrical Engineering and Computer Sciences (EECS) at UC Berkeley advised by Professor Laura Waller. She received a BS in EECS and a BA in Music from UC Berkeley in 2021, graduating as the 150th University Medalist. Her research is in computational imaging, at the intersection of information theory, machine learning, and optics. She is supported by the National Science Foundation Graduate Research Fellowship Program and the Berkeley Fellowship.