WARP: Word-level Adversarial ReProgramming
This work provides a highly parameter-efficient alternative for transfer learning in NLP, which is significant for researchers and practitioners dealing with large language models and limited computational resources.
This paper introduces WARP, an adversarial reprogramming approach that learns task-specific word embeddings to guide pretrained language models for NLP tasks. WARP achieves superior performance on the GLUE benchmark, outperforming methods with up to 1000x more trainable parameters, and also surpasses GPT-3 on two SuperGLUE tasks in a few-shot setting with only 32 training samples.
Transfer learning from pretrained language models recently became the dominant approach for solving many NLP tasks. A common approach to transfer learning for multiple tasks that maximize parameter sharing trains one or more task-specific layers on top of the language model. In this paper, we present an alternative approach based on adversarial reprogramming, which extends earlier work on automatic prompt generation. Adversarial reprogramming attempts to learn task-specific word embeddings that, when concatenated to the input text, instruct the language model to solve the specified task. Using up to 25K trainable parameters per task, this approach outperforms all existing methods with up to 25M trainable parameters on the public leaderboard of the GLUE benchmark. Our method, initialized with task-specific human-readable prompts, also works in a few-shot setting, outperforming GPT-3 on two SuperGLUE tasks with just 32 training samples.