Full Name
Prof. Smita Krishnaswamy
Job Title
Associate Professor
Company
Yale University
Speaker Bio
Smita Krishnaswamy is an Associate Professor of Computer Science and Genetics at Yale University. She is also affiliated with the Program for Applied Mathematics and WTI Institute for NeuroComputation and Machine Intelligence at Yale. Smita's lab works on fundamental deep learning and mathematical machine learning methods for accelerating discovery from biomedical and neuroscientific data. Her work features many methods for generative modeling, graph-based learning, visualization, dynamics modeling and optimal transport as well as multimodal integration of high dimensional data. She has applied her techniques to discovery from cellular, molecular, and imaging data from neuroscience, psychology, stem cell biology, cancer, and immunology. Smita obtained her Ph.D. from the University of Michigan in Computer Science. Smita's work has won several awards including the NSF CAREER Award, Sloan Faculty Fellowship, and Blavatnik Fund for Innovation. Smita teaches deep learning, unsupervised learning, geometry topology in ML and other courses at the intersection of CS and applied math. She also teaches special courses in computational genomics at CGSI (UCLA), CSHL, as well as being a mentor for the yale SUMRY math REU program. This semester, Smita is CRM-Simons Scholar-in-Residence at the Centre Researches Mathematiques at UDEM.
Speaking At
Abstract
Rapid growth of high-dimensional datasets in fields such as single-cell RNA sequencing and spatial genomics has led to unprecedented opportunities for scientific discovery, but it also presents unique computational and statistical challenges. Traditional methods struggle with geometry-aware data generation, interpolation along meaningful trajectories, and transporting populations via feasible paths. To address these issues, we introduce Geometry-Aware Generative Autoencoder (GAGA), a novel framework that combines extensible manifold learning with generative modeling. GAGA constructs a neural network embedding space that respects the intrinsic geometries discovered by manifold learning, enabling the derivation of a Riemannian metric that characterizes the data geometry in the data space. Using this metric, GAGA can uniformly sample points on the manifold, generate points along geodesics, and interpolate between populations across the learned manifold. We demonstrate that GAGA can effectively address all three challenges and significantly improve performance on simulated and real-world datasets.
