Yoav Shechtman
Technion - Israel Institute of Technology
Biophysical Reports Editor

At a cocktail party of non-scientists, how would you explain what you do?

I ruin perfectly good microscopes for a living. My group utilizes the fact that in modern microscopy, there is an extra layer between the human observer and the sample being observed—the computer. This relatively new fact (relative to the history of microscopy) allows for designing novel optical systems that distort images cleverly to encode information in them. This results in images that to the naked eye would seem distorted. The information is then decoded, yielding imaging systems with super-capabilities compared to conventional microscopes. The capabilities we focus on are mainly volumetric and multicolor fluorescence microscopy.

What are you currently working on that excites you?

For the last several years, my group has been using deep learning to solve various microscopy challenges. These include applications in image analysis, such as super-resolution localization microscopy at high density in two dimensions and three dimensions, single-particle diffusion characterization, multicolor imaging, and more; but perhaps the most exciting is neural-net–based optical system design—namely, using algorithms to design the optical system itself such that the measurements are maximally informative. In principle, automated system design frees us from the boundaries of our own human imagination and from falling into local design minima that are sometimes the result of tradition. Of course, algorithmic design has its own limitations, and a human in the loop is still necessary for sanity checks and for some direction of the design process, although we try to let the algorithms “run loose” as much as possible. We are currently pursuing various ways to algorithmically design different application-specific optical systems.