Learning Functions to Study the Benefit of Multitask Learning

TL;DR

This paper investigates how various factors like task relatedness, number of tasks, and samples influence the performance of multitask learning models for sequence labeling, using symbolic regression to uncover performance formulas.

Contribution

It introduces a task simulator and applies symbolic regression to empirically derive formulas linking MTL performance to key factors, extending theoretical insights.

Findings

Performance relates to sqrt(n) and sqrt(T), consistent with prior theory.

Performance also correlates with sqrt(AMI), a measure of task relatedness.

Formulas provide a quantitative understanding of factors affecting MTL outcomes.

Abstract

We study and quantify the generalization patterns of multitask learning (MTL) models for sequence labeling tasks. MTL models are trained to optimize a set of related tasks jointly. Although multitask learning has achieved improved performance in some problems, there are also tasks that lose performance when trained together. These mixed results motivate us to study the factors that impact the performance of MTL models. We note that theoretical bounds and convergence rates for MTL models exist, but they rely on strong assumptions such as task relatedness and the use of balanced datasets. To remedy these limitations, we propose the creation of a task simulator and the use of Symbolic Regression to learn expressions relating model performance to possible factors of influence. For MTL, we study the model performance against the number of tasks (T), the number of samples per task (n) and the task relatedness measured by the adjusted mutual information (AMI). In our experiments, we could empirically find formulas relating model performance with factors of sqrt(n), sqrt(T), which are equivalent to sound mathematical proofs in Maurer[2016], and we went beyond by discovering that performance relates to a factor of sqrt(AMI).