Dissertation Talk: Time-Warp Manifolds

Event details Date: Friday, July 31, 2026 Time: 1:00 PM to 1:00 PM Location: Online Type: Lecture / Workshop Audience: Faculty,Students About this event This dissertation addresses the challenge of creating stable and interpretable low-dimensional embeddings of multidimensional time-series data. The goal of an embedding is to represent the elements of a high-dimensional space in a lower-dimensional space while approximately preserving the pairwise distances between elements, a problem known as distortion minimization. How well the distortion can be minimized, and how much the result means, depends on the choice of distance function. While numerous distance functions can be adapted for time-series embeddings, many are unstable under perturbation or discard the timing relationships that carry much of a signal’s meaning. This work studies a class of distances known as “time-warp” distances, defined as the optimal values of optimization problems that align two signals by warping time. It develops the numerical and theoretical aspects of the stable recovery of time-warp functions, both individually and in populations, and offers a geometric and topological treatment of the embeddings built from them. Official event details: https://events.berkeley.edu/eecs/event/324437-dissertation-talk-time-warp-manifolds