Multi-task Learning over Graph Structures

TL;DR

This paper introduces a flexible graph-based multi-task learning framework for neural sequence models that enables dynamic task relationship learning, leading to improved performance and interpretability across text classification and sequence labeling tasks.

Contribution

It proposes a novel graph multi-task learning architecture that allows tasks to communicate dynamically, unlike previous fixed-structure methods.

Findings

Models outperform baseline methods in text classification and sequence labeling.

The approach learns interpretable patterns across tasks.

Demonstrates effective transfer learning capabilities.

Abstract

We present two architectures for multi-task learning with neural sequence models. Our approach allows the relationships between different tasks to be learned dynamically, rather than using an ad-hoc pre-defined structure as in previous work. We adopt the idea from message-passing graph neural networks and propose a general \textbf{graph multi-task learning} framework in which different tasks can communicate with each other in an effective and interpretable way. We conduct extensive experiments in text classification and sequence labeling to evaluate our approach on multi-task learning and transfer learning. The empirical results show that our models not only outperform competitive baselines but also learn interpretable and transferable patterns across tasks.