WARP: Word-level Adversarial ReProgramming

By Karen Hambardzumyan, Hrant Khachatrian, Jonathan May. 2021

380 citations recorded by the source corpus.

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.

Subfields: Artificial Intelligence, Safety, Robustness & Interpretability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Safety, Robustness & Interpretability

Citations in the corpus

The Power of Scale for Parameter-Efficient Prompt Tuning (2021, methodology): “While we developed our method concurrently with Li and Liang (2021) and Hambardzumyan et al. (2021), we are the ﬁrst to show that prompt tuning alone (with no intermediate-layer preﬁxes or task-speciﬁc output layers) is sufﬁcient to be competitive with model tuning.”
Prefix-Tuning: Optimizing Continuous Prompts for Generation (2021, background): “Hambardzumyan et al. (2021) learns task-specific embeddings that adapts BERT for sentiment classification.”
Pre-train, Prompt, and Predict: A Systematic Survey of Prompting Methods in Natural Language Processing (2021, methodology): “[40] assign a virtual token for each class label and optimize the token embedding for each class together with prompt token embeddings.”
LoRA: Low-Rank Adaptation of Large Language Models (2021, background): “More recently, [14, 19, 21, 25] proposed optimizing the input word embeddings in lieu of fine-tuning, akin to a continuous and differentiable generalization of prompt engineering.”
Compacter: Efficient Low-Rank Hypercomplex Adapter Layers (2021, neutral): “Soft prompt methods treat prompts as trainable continuous parameters, which are prepended to the inputs at the input layer or intermediate layers [10, 11, 12].”
Surface Form Competition: Why the Highest Probability Answer Isn’t Always Right (2021, background): “Contemporary work has shown it is possible to get smaller models to exhibit few-shot learning behavior using fine-tuning (Hambardzumyan et al., 2021; Gao et al., 2020; Schick and Schütze, 2020a,b,c; Shin et al., 2020), an intermediate learning phase (Ye et al.”
Pre-trained models for natural language processing: A survey (2020, background): “Continuous WARP [73],Prefix-Tuning [74],P-Tuning [75]”
What Changes Can Large-scale Language Models Bring? Intensive Study on HyperCLOVA: Billions-scale Korean Generative Pretrained Transformers (2021, background): “Recent work (Li and Liang, 2021; Hambardzumyan et al., 2021; Liu et al., 2021b; Lester et al., 2021) proposed optimizing the contextualized token spaces without ﬁne-tuning the main LM parameters.”
SPoT: Better Frozen Model Adaptation through Soft Prompt Transfer (2021, methodology): “Compared to WARP (Hambardzumyan et al., 2021), our SPOT approach tunes 16×more parameters (410K vs. 25K), and benefits from 50× more frozen parameters.”
PPT: Pre-trained Prompt Tuning for Few-shot Learning (2021, background): “To overcome the shortcomings of discrete spaces, Li and Liang (2021); Liu et al. (2021); Han et al. (2021b); Hambardzumyan et al. (2021); Zhong et al. (2021b) explore to combine hard prompts and soft prompts.”
Why Do Pretrained Language Models Help in Downstream Tasks? An Analysis of Head and Prompt Tuning (2021, positive): “Prompt tuning [31, 19, 9, 21] optimizes a task-specific “prompt” that is concatenated to the model input.”
On Robust Prefix-Tuning for Text Classification (2022, methodology): “Our framework is also applicable to other soft prompt-based tuning methods (Qin & Eisner, 2021; Hambardzumyan et al., 2021; Lester et al., 2021; Cho et al., 2021; Tsimpoukelli et al., 2021) by recording the activations of correctly classiﬁed training data, constructing canonical manifolds for the…”
Knowledgeable Prompt-tuning: Incorporating Knowledge into Prompt Verbalizer for Text Classification (2021, neutral): “The soft verbalizer is proposed by WARP (Hambardzumyan et al., 2021).”
Prompting PaLM for Translation: Assessing Strategies and Performance (2022, background): “, 2021), activations (Li and Liang, 2021; Hambardzumyan et al., 2021), or attention weights (Liu et al.”
Tuning Language Models as Training Data Generators for Augmentation-Enhanced Few-Shot Learning (2022, methodology): “Later developments in prompt-based methods replace the manual design of prompt templates with automatic search or learning (Cui et al., 2022; Hambardzumyan et al., 2021; Lester et al., 2021; Liu et al., 2021b; Zhang et al., 2022; Zhong et al., 2021).”
Exploring Low-dimensional Intrinsic Task Subspace via Prompt Tuning (2021, methodology): “Following this line of study, the prompts are extended from tokens in the vocabularies to trainable embeddings, i.e., soft prompts (Li and Liang, 2021; Hambardzumyan et al., 2021; Zhong et al., 2021; Qin and Eisner, 2021; Liu et al., 2021b; Lester et al., 2021).”
KnowPrompt: Knowledge-aware Prompt-tuning with Synergistic Optimization for Relation Extraction (2021, background): “Recently, some continuous prompts have also been proposed [21, 25, 34, 36], which focus on utilizing learnable continuous embeddings as prompt templates rather than label words.”