Spectral Experts for Estimating Mixtures of Linear Regressions
read the original abstract
Discriminative latent-variable models are typically learned using EM or gradient-based optimization, which suffer from local optima. In this paper, we develop a new computationally efficient and provably consistent estimator for a mixture of linear regressions, a simple instance of a discriminative latent-variable model. Our approach relies on a low-rank linear regression to recover a symmetric tensor, which can be factorized into the parameters using a tensor power method. We prove rates of convergence for our estimator and provide an empirical evaluation illustrating its strengths relative to local optimization (EM).
This paper has not been read by Pith yet.
Forward citations
Cited by 2 Pith papers
-
Linking COPD Prevalence with Income Distribution: A Spatial Heterogeneous Compositional Regression via Geographically Weighted Penalized Approach
A new penalized geographically weighted compositional regression detects both contiguous and non-contiguous spatial clusters with shared effects when linking income distributions to COPD prevalence.
-
Linking COPD Prevalence with Income Distribution: A Spatial Heterogeneous Compositional Regression via Geographically Weighted Penalized Approach
A new geographically weighted penalized compositional regression model with pairwise fusion penalty is proposed to handle spatial heterogeneity and compositional covariates, demonstrated on U.S. income and COPD data.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.