Expert-guided Regularization via Distance Metric Learning

TL;DR

This paper introduces DMLreg, a novel regularization method that incorporates expert knowledge into high-dimensional linear models by learning a Mahalanobis distance metric from expert comparisons, improving prediction accuracy.

Contribution

It proposes a new regularization technique that leverages domain expert input via distance metric learning to enhance high-dimensional model performance.

Findings

DMLreg improves prediction accuracy with knowledgeable experts.

The learned distance metric effectively encodes prior feature importance.

Experimental results validate the approach on simulated data.

Abstract

High-dimensional prediction is a challenging problem setting for traditional statistical models. Although regularization improves model performance in high dimensions, it does not sufficiently leverage knowledge on feature importances held by domain experts. As an alternative to standard regularization techniques, we propose Distance Metric Learning Regularization (DMLreg), an approach for eliciting prior knowledge from domain experts and integrating that knowledge into a regularized linear model. First, we learn a Mahalanobis distance metric between observations from pairwise similarity comparisons provided by an expert. Then, we use the learned distance metric to place prior distributions on coefficients in a linear model. Through experimental results on a simulated high-dimensional prediction problem, we show that DMLreg leads to improvements in model performance when the domain expert is knowledgeable.