Bandits for Learning to Explain from Explanations

TL;DR

This paper presents Explearn, an online algorithm using Gaussian Process-based contextual bandits to jointly learn predictions and explanations, offering controlled generalization and theoretical convergence guarantees.

Contribution

Introduction of Explearn, a novel GP-based contextual bandit algorithm that learns to generate explanations alongside predictions with provable convergence.

Findings

Initial experiments show promising results

GPs enable flexible explanation modeling

The approach guarantees convergence with high probability

Abstract

We introduce Explearn, an online algorithm that learns to jointly output predictions and explanations for those predictions. Explearn leverages Gaussian Processes (GP)-based contextual bandits. This brings two key benefits. First, GPs naturally capture different kinds of explanations and enable the system designer to control how explanations generalize across the space by virtue of choosing a suitable kernel. Second, Explearn builds on recent results in contextual bandits which guarantee convergence with high probability. Our initial experiments hint at the promise of the approach.