Computational Methods Enable Extraordinary Imaging Systems

Speaker: Dr. Jingtian Hu
Affiliation: Northwestern University

Via Zoom only:    https://ucla.zoom.us/j/96100391649?pwd=bFcvdFJVbU1nY284akhhcjF1L1d5QT09

Zoom Meeting ID: 961 0039 1649

Password: 466812

Abstract:  Artificial intelligence has enabled unprecedented opportunities for scientists ranging from experimental design to data analysis. In the field of photonics, both traditional evolutionary algorithms and data-driven machine learning approaches has contributed to novel photonics designs. Machine learning methods also start to play a key role in processing optical information from spectroscopy and imaging of photonic nanostructures. This presentation describes how computational approaches combined with nanoscience techniques enable (1) novel designs of metasurface imaging devices and (2) analysis methods that can extract subwavelength structural information from optical microscopy.

Lattice-resonance metalenses for reconfigurable imaging: A reconfigurable metalens system that can image at visible wavelengths has been demonstrated based on arrays of coupled plasmonic nanoparticles. These lenses manipulated the wavefront and focused light by exciting surface lattice resonances that were tuned by patterned polymer blocks on single- particle sites. Predictive design of the dielectric nano-blocks was performed using an evolutionary algorithm to create a range of 3D focusing responses. We demonstrated a simple technique for erasing and writing the polymer nanostructures on the metal nanoparticle arrays in a single step using solvent assisted nanoscale embossing. This reconfigurable materials platform enables tunable focusing with diffraction-limited resolution and offers prospects for highly adaptive, compact-imaging.

Deep learning algorithms for tracking rotational dynamics: An imaging platform is reported based on deep-learning algorithms and differential interference contrast (DIC) microscopy for tracking the 3D rotations of nanoparticle optical probes. To establish the deep-learning models, we constructed large image datasets of anisotropic nanoparticles with labeled orientations. In all imaging environments, the optimized models could predict the particle orientations from their DIC images with an accuracy limited only by instrumental approach. Lastly, we demonstrated that the deep-learning model could achieve 3D orientation tracking with in-plane and out-of-plane rotations determined simultaneously.

Biography:   Jingtian Hu received his B.S. degree in Materials Science and Engineering from University of Illinois at Urbana-Champaign in 2013. He received a Ph.D. degree in Materials Science and Engineering at Northwestern University advised by Prof. Teri W. Odom in March 2019. His research focuses on predictive design of nanophotonic devices by evolutionary algorithms and machine-learning particle-tracking methods for nanoscale nanoparticle dynamics. He has participated in Hierarchical Materials Cluster Program in 2015 and Predictive Science and Engineering Design Cluster Program in 2016 and has been recognized as IIN Outstanding Researcher in 2017, John E. Hilliard Symposium Fellow in 2019 and China Scholarship Council Award Winner in 2020.

For more information, contact Prof. Aydogan Ozcan (ozcan@g.ucla.edu)

Date/Time:
Date(s) - Jan 06, 2021
4:00 pm - 5:30 pm

Location:
Via Zoom Only
No location, Los Angeles
Map Unavailable