Black-Box Prompt Tuning With Subspace Learning
- 7 June 2024
- journal article
- research article
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE/ACM Transactions on Audio, Speech, and Language Processing
- Vol. 32 (23299290) , 3002-3013
Abstract
Black-box prompt tuning employs derivative-free optimization algorithms to learn prompts within low-dimensional subspaces rather than back-propagating through the network of Large Language Models (LLMs). Recent studies reveal that black-box prompt tuning lacks versatility across tasks and LLMs, which we believe is related to the suboptimal choice of subspaces. In this paper, we introduce B lack-box prompt tuning with S ubspace L earning (BSL) to enhance the versatility of black-box prompt tuning. Based on the assumption that nearly optimal prompts for similar tasks reside in a common subspace, we propose identifying such subspaces through meta-learning on a collection of similar source tasks. Consequently, for a target task that shares similarities with the source tasks, we expect that optimizing within the identified subspace can yield a prompt that performs well on the target task. Experimental results confirm that our BSL framework consistently achieves competitive performance across various downstream tasks and LLMs.Keywords
All Related Versions
Funding Information
- National Key R&D Program of China (2022ZD0160502)
- National Natural Science Foundation of China (61925601, 62006138)
- National Social Science Fund of China (20&ZD279)
This publication has 22 references indexed in Scilit:
- P-Tuning: Prompt Tuning Can Be Comparable to Fine-tuning Across Scales and Tasks2022
- Making Pre-trained Language Models Better Few-shot Learners2021
- The E2E Dataset: New Challenges For End-to-End Generation2017
- Development of a benchmark corpus to support the automatic extraction of drug-related adverse effects from medical case reportsJournal of Biomedical Informatics, 2012
- Introduction to Derivative-Free Optimization2009
- The PASCAL Recognising Textual Entailment Challenge2006
- Annotating Expressions of Opinions and Emotions in LanguageLanguage Resources and Evaluation, 2005
- Mining and summarizing customer reviews2004
- Reducing the Time Complexity of the Derandomized Evolution Strategy with Covariance Matrix Adaptation (CMA-ES)Evolutionary Computation, 2003
- Completely Derandomized Self-Adaptation in Evolution StrategiesEvolutionary Computation, 2001