Unsupervised Cross-Task Generalization via Retrieval Augmentation
It addresses the challenge of enabling language models to generalize to unseen tasks without supervision, which is incremental as it builds on existing multi-task models like T0 and FLAN.
The paper tackles the problem of improving cross-task generalization for multi-task language models in an unsupervised setting by proposing ReCross, a retrieval-augmentation method that uses unlabelled examples to retrieve and update with upstream data, resulting in significant performance gains over baselines.
Humans can perform unseen tasks by recalling relevant skills acquired previously and then generalizing them to the target tasks, even if there is no supervision at all. In this paper, we aim to improve this kind of cross-task generalization ability of massive multi-task language models, such as T0 and FLAN, in an unsupervised setting. We propose a retrieval-augmentation method named ReCross that takes a few unlabelled examples as queries to retrieve a small subset of upstream data and uses them to update the multi-task model for better generalization. ReCross is a straightforward yet effective retrieval method that combines both efficient dense retrieval and effective pair-wise reranking. Our results and analysis show that it significantly outperforms both non-retrieval methods and other baseline methods.