Improved Estimation in Time Varying Models

TL;DR

This paper introduces a variance reduction method for time-varying models by learning a transformed space and task-driven bases, demonstrated through synthetic and EEG classification data.

Contribution

It proposes a novel approach for variance reduction in locally adapted models by learning a transformed space and bases, including task-driven extensions.

Findings

Effective variance reduction in synthetic data experiments

Improved EEG classification performance

Learned bases capture meaningful network structures

Abstract

Locally adapted parameterizations of a model (such as locally weighted regression) are expressive but often suffer from high variance. We describe an approach for reducing the variance, based on the idea of estimating simultaneously a transformed space for the model, as well as locally adapted parameterizations in this new space. We present a new problem formulation that captures this idea and illustrate it in the important context of time varying models. We develop an algorithm for learning a set of bases for approximating a time varying sparse network; each learned basis constitutes an archetypal sparse network structure. We also provide an extension for learning task-driven bases. We present empirical results on synthetic data sets, as well as on a BCI EEG classification task.