TaskWeb: Selecting Better Source Tasks for Multi-task NLP
This work addresses the challenge of task selection for multi-task learning in NLP, offering a practical solution for researchers and practitioners, though it is incremental as it builds on existing multi-task frameworks.
The paper tackles the problem of selecting source tasks for multi-task NLP by introducing TaskWeb, a benchmark of pairwise task transfers, and TaskShop, a method that uses this benchmark to choose helpful training tasks. The result is a 10% improvement in overall rankings and a 38% improvement in top-k precision for source tasks, with smaller training sets boosting zero-shot performance by at least 4.3% across 11 target tasks.
Recent work in NLP has shown promising results in training models on large amounts of tasks to achieve better generalization. However, it is not well-understood how tasks are related, and how helpful training tasks can be chosen for a new task. In this work, we investigate whether knowing task relationships via pairwise task transfer improves choosing one or more source tasks that help to learn a new target task. We provide TaskWeb, a large-scale benchmark of pairwise task transfers for 22 NLP tasks using three different model types, sizes, and adaptation methods, spanning about 25,000 experiments. Then, we design a new method TaskShop based on our analysis of TaskWeb. TaskShop uses TaskWeb to estimate the benefit of using a source task for learning a new target task, and to choose a subset of helpful training tasks for multi-task training. Our method improves overall rankings and top-k precision of source tasks by 10% and 38%, respectively. We also use TaskShop to build much smaller multi-task training sets that improve zero-shot performances across 11 different target tasks by at least 4.3%.