Provable Tensor Methods for Learning Mixtures of Generalized Linear\n Models
Year: 2017
Type: article
Abstract: We consider the problem of learning mixtures of generalized linear models\n(GLM) which arise in classification and regression problems. Typical learning\napproaches such as expectation maximization (EM) or variational Bayes can get\nstuck in spurious local optima. In contrast, we present a tensor decomposition\nmethod which is guaranteed to correctly recover the parameters. The key insight\nis to employ certain feature transformations of the input, which depend on the\ninput generative model. Specifically, we employ score function tensors of the\ninput and compute their cross-correlation with the response variable. We\nestablish that the decomposition of this tensor consistently recovers the\nparameters, under mild non-degeneracy conditions. We demonstrate that the\ncomputational and sample complexity of our method is a low order polynomial of\nthe input and the latent dimensions.
Authors: Sedghi, H, Janzamin, M, Anandkumar, A
Institution: University of California, Irvine
Language: English
Subfield: Computational Mathematics
Field: Mathematics
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W4293195944
