Vector-Quantized Input-Contextualized Soft Prompts for Natural Language Understanding
This addresses the problem of poor generalization in prompt tuning for natural language understanding tasks, offering a more robust method for researchers and practitioners, though it is incremental as it builds on existing soft prompt tuning frameworks.
The paper tackled the limitation of fixed soft prompts in prompt tuning by proposing Vector-quantized Input-contextualized Prompts (VIP), which adapts prompts to input-specific contexts and quantizes them, resulting in an average improvement of 1.19% over baseline on various language understanding tasks and better generalization in out-of-domain and multi-task settings.
Prompt Tuning has been largely successful as a parameter-efficient method of conditioning large-scale pre-trained language models to perform downstream tasks. Thus far, soft prompt tuning learns a fixed set of task-specific continuous vectors, i.e., soft tokens that remain static across the task samples. A fixed prompt, however, may not generalize well to the diverse kinds of inputs the task comprises. In order to address this, we propose Vector-quantized Input-contextualized Prompts (VIP) as an extension to the soft prompt tuning framework. VIP particularly focuses on two aspects -- contextual prompts that learns input-specific contextualization of the soft prompt tokens through a small-scale sentence encoder and quantized prompts that maps the contextualized prompts to a set of learnable codebook vectors through a Vector quantization network. On various language understanding tasks like SuperGLUE, QA, Relation classification, NER and NLI, VIP outperforms the soft prompt tuning (PT) baseline by an average margin of 1.19%. Further, our generalization studies show that VIP learns more robust prompt representations, surpassing PT by a margin of 0.6% - 5.3% on Out-of-domain QA and NLI tasks respectively, and by 0.75% on Multi-Task setup over 4 tasks spanning across 12 domains.