Timezone: »

 
Poster
Iterative Least Trimmed Squares for Mixed Linear Regression
Yanyao Shen · Sujay Sanghavi

Tue Dec 10 05:30 PM -- 07:30 PM (PST) @ East Exhibition Hall B + C #50
in Algorithms -- Regression »

Given a linear regression setting, Iterative Least Trimmed Squares (ILTS) involves alternating between (a) selecting the subset of samples with lowest current loss, and (b) re-fitting the linear model only on that subset. Both steps are very fast and simple. In this paper, we analyze ILTS in the setting of mixed linear regression with corruptions (MLR-C). We first establish deterministic conditions (on the features etc.) under which the ILTS iterate converges linearly to the closest mixture component. We also provide a global algorithm that uses ILTS as a subroutine, to fully solve mixed linear regressions with corruptions. We then evaluate it for the widely studied setting of isotropic Gaussian features, and establish that we match or better existing results in terms of sample complexity. Finally, we provide an ODE analysis for a gradient-descent variant of ILTS that has optimal time complexity. Our results provide initial theoretical evidence that iteratively fitting to the best subset of samples -- a potentially widely applicable idea -- can provably provide state of the art performance in bad training data settings.

Keywords: Regression · Non-Convex Optimization · Algorithms · Algorithms -> Large Scale Learning; Optimization
[ Paper]  [ Poster] 

Author Information

Yanyao Shen (UT Austin)
Sujay Sanghavi (UT-Austin)

More from the Same Authors