This is an online seminar. Registration is required.
【Tensor Learning Team】
【Date】2024/April/15(Mon) 15:00-16:00(JST)
*【Speaker】Zerui Tao, Tokyo University of Agriculture and Technology *
Title: Flexible Probabilistic Tensor Decomposition for Multiway Data
Abstract:
Tensor decomposition serves as an important tool to analyze multiway data.However, traditional tensor decompositions encounter two significant challenges when modeling high-dimensional multiway data. Firstly, the generative process of traditional models is typically restricted to multi-linear structures and simple distributions, which are not sufficient to represent real-world data.
Secondly, computing posterior distributions for high-dimensional multiway data poses computational challenges. In this talk, I will discuss our recent research on flexible probabilistic tensor decompositions, which expand traditional tensor decomposition to nonlinear structures and unknown distributions. To infer the posteriors of these
models, we establish efficient Gibbs sampler, variational inference and noise-contrastive estimation algorithms. The proposed models exhibit promising performance on tensor completion tasks.
