12:30–1:30 pm Searle 240A
Tuesday, February 20, 2024, at 12:30 PM, in Searle 240A (location change this quarter), 5735 South Ellis Avenue
CAM & Stats Students Seminar
John Hood, Department of Statistics, The University of Chicago
"The Allocore Tensor Decomposition"
Abstract
We introduce Allocore, a new form of probabilistic tensor decomposition. Allocore is a Tucker decomposition where the L0-norm of the core tensor is constrained to be at most Q. While the user dictates the total budget Q, the locations and values of the non-zero elements are latent variables allocated across the core tensor during inference. Allocore—i.e., allocated L0-constrained core— enjoys both the computational tractability of CP decomposition and the qualitatively appealing latent structure of Tucker. In a suite of real-data experiments, we demonstrate that Allocore typically requires only tiny fractions (e.g., 1%) of the full core to achieve the same results as full Tucker at only a correspondingly small fraction of the cost.