Multi-task Learning over Graph Structures
arXiv:1811.10211
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.
arXiv commentAccepted by AAAI 2019
References in corpus (9)
- Semi-Supervised Classification with Graph Convolutional Networks
- ADADELTA: An Adaptive Learning Rate Method
- Neural Message Passing for Quantum Chemistry
- A Convolutional Neural Network for Modelling Sentences
- Multi-Task Cross-Lingual Sequence Tagging from Scratch
- Long Short-Term Memory-Networks for Machine Reading
- Deep Multitask Learning for Semantic Dependency Parsing
- Exploiting Multi-typed Treebanks for Parsing with Deep Multi-task Learning
- Meta-Learning Multi-task Communication