MetaPrompting: Learning to Learn Better Prompts
This addresses a key bottleneck for researchers and practitioners in few-shot NLP by enabling more effective and efficient adaptation of soft prompts to new tasks.
The paper tackles the problem of soft prompt initialization in few-shot NLP by proposing MetaPrompting, which uses meta-learning to automatically find better initializations, resulting in over 6 points accuracy improvement in 1-shot settings across four datasets.
Prompting method is regarded as one of the crucial progress for few-shot nature language processing. Recent research on prompting moves from discrete tokens based ``hard prompts'' to continuous ``soft prompts'', which employ learnable vectors as pseudo prompt tokens and achieve better performance. Though showing promising prospects, these soft-prompting methods are observed to rely heavily on good initialization to take effect. Unfortunately, obtaining a perfect initialization for soft prompts requires understanding of inner language models working and elaborate design, which is no easy task and has to restart from scratch for each new task. To remedy this, we propose a generalized soft prompting method called MetaPrompting, which adopts the well-recognized model-agnostic meta-learning algorithm to automatically find better prompt initialization that facilitates fast adaptation to new prompting tasks.Extensive experiments show MetaPrompting tackles soft prompt initialization problem and brings significant improvement on four different datasets (over 6 points improvement in accuracy for 1-shot setting), achieving new state-of-the-art performance.