“Computational image understanding via statistical inference and machine learning: spatial-temporal structural biology and computer vision”
Thursday, Feb. 28 @ 11:45 am
Challenging interdisciplinary applications inspire new methodological developments in data processing. This talk presents methodologies ranging from statistical inference based on analytical models to machine learning based on deep neural networks, motivated by understanding large scale real-world data.
The primary focus of the talk is my work on recovering 3-D spatial structure and temporal dynamics of nanoscale biological particles (e.g., viruses and ribosomes) directly from large sets of cryo electron microscopy data, where we seek to provide quantitative evidence for scientific hypotheses. With a proposed statistical framework incorporating the continuous heterogeneity among the imaged particles, we develop a generative mechanical model to provide sparse and analytical parametrization of the stochastic description of particle structure. This work contributes a novel model-based solution for the open problem of large covariance estimation, as well as a systematic way to incorporate a fourth dimension to the concept of 3D reconstruction. The talk also presents my work on concept learning, as an example of a method with a different perspective on data understanding based on deep generative models. The talk concludes by discussing how physics-based statistical inference and deep learning might be fused to provide a hybrid framework that incorporates the strengths of both components in the context of interdisciplinary problems, especially targeting challenges such as data scarcity, model resilience, robustness and interpretability of the solutions.
Yunye Gong is a Ph.D. candidate in Electrical and Computer Engineering at Cornell University working with Prof. Peter C. Doerschuk. She received her Bachelor of Science degree in Electrical Engineering from the University of Illinois at Urbana-Champaign (UIUC) in 2013. She received the Google Ph.D. Fellowship in Machine Learning in 2017. Her research interests lie in broadening the horizon of computational signal and image understanding, targeting methodological challenges that arise from real-world problems of importance to diverse application domains in science and engineering.