WARP: Word-level Adversarial ReProgramming

Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing, pages 4921–4933 August 1–6, 2021. ©2021 Association for Computational Linguistics 4921 WARP: Word-level Adversarial ReProgramming Karen Hambardzumyan1, Hrant Khachatrian1,2, Jonathan May3 1YerevaNN, 2Yerevan State University, 3Information Sciences Institute, University of Southern California mahnerak@yerevann.com, hrant@yerevann.com, jonmay@isi.edu Abstract Transfer learning from pretrained language models recently became the dominant ap- proach for solving many NLP tasks. A com- mon 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 re- programming, which extends earlier work on automatic prompt generation. Adversarial re- programming 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 leader- board of the GLUE benchmark. Our method, initialized with task-specific human-readable prompts, also works in a few-shot setting, out- performing GPT-3 on two SuperGLUE tasks with just 32 training samples. 1 Introduction Language model pretraining has had a tremendous impact on solving many natural language process- ing tasks (Peters et al., 2018; Radford et al., 2018; Devlin et al., 2019; Liu et al., 2019). The most popular two approaches take a pretrained model and use a straightforward supervised learning ob- jective. In the first approach, the parameters of the language model are frozen and a task-specific head is trained on top of them (Peters et al., 2018). The second approach fine-tunes all model param- eters (Radford et al., 2018). The latter can some- times yield better results (Peters et al., 2019), while the first one usually offers better stability for smaller datasets. The approach based on frozen features does not require storing task-specific lan- guage models. A recent alternative is based on so called adapters (Houlsby et al., 2019; Pfeiffer et al., 2021), a technique that adds new weights at every layer of the pretrained language model while the original parameters are kept frozen. This enables a smaller set of task-specific parameters while achieving results comparable to the fine-tuning ap- proach. Another approach of leveraging pretrained lan- guage models for downstream tasks, introduced by Radford et al. (2019), provides “task descrip- tions” without using any labeled examples. GPT- 3 (Brown et al., 2020) demonstrates impressive few-shot learning performance with priming: by providing the language model a few inputs and outputs (“analogies”) as a context. The language model contextually “learns” from these examples and outputs the answer with a single forward pass without any trainable parameters. These methods, however, require huge language models (1.5B and 175B parameters, respectively). The success of task reformulation-based ap- proaches suggest that language models are capa- ble of solving various natural language processing tasks given a well-crafted prompt. We hypothesize that it is possible to …

Content truncated to 500 words