References & Citations
Mathematics > Numerical Analysis
Title: Learning Diagonal Gaussian Mixture Models and Incomplete Tensor Decompositions
(Submitted on 8 Feb 2021 (v1), last revised 8 Jun 2021 (this version, v2))
Abstract: This paper studies how to learn parameters in diagonal Gaussian mixture models. The problem can be formulated as computing incomplete symmetric tensor decompositions. We use generating polynomials to compute incomplete symmetric tensor decompositions and approximations. Then the tensor approximation method is used to learn diagonal Gaussian mixture models. We also do the stability analysis. When the first and third order moments are sufficiently accurate, we show that the obtained parameters for the Gaussian mixture models are also highly accurate. Numerical experiments are also provided.
Submission history
From: Jiawang Nie [view email][v1] Mon, 8 Feb 2021 19:49:32 GMT (923kb,D)
[v2] Tue, 8 Jun 2021 21:58:10 GMT (924kb,D)
Link back to: arXiv, form interface, contact.