“A kernel Stein test of goodness of fit for sequential models

Heishiro Kanagawa (Newcastle University) -
Data Science

Date: -
Location: Eurecom

Abstract: I will discuss a new goodness-of-fit measure for probability densities modeling observations with varying dimensionality, such as text documents of differing lengths or variable-length sequences. The proposed measure is an instance of the kernel Stein discrepancy (KSD), which has been used to construct goodness-of-fit tests for unnormalized densities. The KSD is defined by its Stein operator: current operators used in testing apply to fixed-dimensional spaces. In this talk, we will see the KSD extended to the variable-dimension setting by identifying appropriate Stein operators, and a novel KSD goodness-of-fit test will be constructed. As with the previous variants, the proposed KSD does not require the density to be normalized, allowing the evaluation of a large class of models, including Markov random field models and output-constrained generative models. Reference: Baum, J., Kanagawa, H., & Gretton, A. (2023). A kernel Stein test of goodness of fit for sequential models, ICML 2023 https://arxiv.org/abs/2210.10741 Bio: Heishiro is a postdoctoral researcher at Newcastle University in the group of Chris Oates. He obtained his PhD from the Gatsby Computational Neuroscience Unit at UCL under the supervision of Arthur Gretton. He received BSc and MSc from Tokyo Institute of Technology, where he worked with Taiji Suzuki.