Join Zoom Meeting
?Geometric and Topological Approaches toRepresentation Learning in Biomedical Data?
Abstract: In this talk, I will overview datageometric and topological approaches to understanding the shape and structureof the data. First, we show how diffusion geometry and deep learning can beused to obtain useful representations of the data that enable denoising,dimensionality reduction, and factor analysis of the data. Next we will showhow to learn dynamics from static snapshot data by using a manifold-regularizedneural ODE-based optimal transport (TrajectoryNet). Finally, we cover a novelapproachto combine diffusion geometry with topology to extract multi-granularfeatures from the data (Diffusion Condensation and Multiscale PHATE).
Bio: Smita Krishnaswamyis an Associate Professor inthe department of Genetics at the Yale School of Medicine and Department ofComputer Science in the Yale School of Applied Science and Engineering and acore member of the Program in Applied Mathematics. She is also affiliated withtheYale Center for Biomedical Data Science, Yale Cancer Center, and PrograminInterdisciplinary Neuroscience. Smita’s research focuses on developingunsupervised machine learning methods (especially graph signal processing anddeep-learning) to denoise, impute, visualize and extract structure, patternsand relationships from big, high throughput, high dimensional biomedical data.Her methods have been applied variety of datasets from many systems includingembryoid body differentiation, zebrafish development, theepithelial-to-mesenchymal transition in breast cancer, lung cancerimmunotherapy, infectious disease data, gut microbiome data and patient data.
Smita teaches three courses: Machine Learning for Biology(Fall), Deep Learning Theory and applications (spring), Advanced Topics in MachineLearning & Data Mining (Spring). She completed her postdoctoral training atColumbia University in the systems biology department where she focused onlearning computational models of cellular signaling from single-cell masscytometrydata. She was trained as a computer scientist with a Ph.D. from theUniversity of Michigan’s EECS department where her research focused onalgorithms for automated synthesis and probabilistic verification of nanoscalelogic circuits. Following her time in Michigan, Smita spent 2 years at IBM’s TJWatson Research Center as a researcher in the systems division where she workedon automated bug finding and error correction in logic.