Bayesian Multitask Learning with Latent Hierarchies

TL;DR

This paper introduces a hierarchical Bayesian model for multitask learning that captures latent relationships between tasks, effectively sharing information based on the task context, and demonstrates superior performance on real-world datasets.

Contribution

The paper proposes a novel hierarchical Bayesian framework that unifies various multitask learning models and adapts sharing strategies for different task relationships.

Findings

Model subsumes several existing multitask learning approaches.

Performs well on three real-world datasets.

Effectively captures latent hierarchical task relationships.

Abstract

We learn multiple hypotheses for related tasks under a latent hierarchical relationship between tasks. We exploit the intuition that for domain adaptation, we wish to share classifier structure, but for multitask learning, we wish to share covariance structure. Our hierarchical model is seen to subsume several previously proposed multitask learning models and performs well on three distinct real-world data sets.