Robust Prompt Optimization for Large Language Models Against Distribution Shifts

被引:0
|
作者
Li, Moxin [1 ]
Wang, Wenjie [1 ]
Feng, Fuli [2 ,3 ]
Cao, Yixin [4 ]
Zhang, Jizhi [2 ]
Chua, Tat-Seng [1 ]
机构
[1] Natl Univ Singapore, Singapore, Singapore
[2] Univ Sci & Technol China, Hefei, Peoples R China
[3] Inst Dataspace, Hefei, Anhui, Peoples R China
[4] Singapore Management Univ, Singapore, Singapore
关键词
D O I
暂无
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Large Language Model (LLM) has demonstrated significant ability in various Natural Language Processing tasks. However, their effectiveness is highly dependent on the phrasing of the task prompt, leading to research on automatic prompt optimization using labeled task data. We reveal that these prompt optimization techniques are vulnerable to distribution shifts such as subpopulation shifts, which are common for LLMs in real-world scenarios such as customer reviews analysis. In this light, we propose a new problem of robust prompt optimization for LLMs against distribution shifts, which requires the prompt optimized over the labeled source group can simultaneously generalize to an unlabeled target group. To solve this problem, we propose Generalized Prompt Optimization framework, which incorporates the unlabeled data from the target group into prompt optimization. Extensive experimental results demonstrate the effectiveness of the proposed framework with significant performance improvement on the target group and comparable performance on the source group.
引用
收藏
页码:1539 / 1554
页数:16
相关论文
共 50 条
  • [31] Quantifying the Performance of Adversarial Training on Language Models with Distribution Shifts
    Omar, Marwan
    Choi, Soohyeon
    Nyang, DaeHun
    Mohaisen, David
    CYSSS'22: PROCEEDINGS OF THE 1ST WORKSHOP ON CYBERSECURITY AND SOCIAL SCIENCES, 2022, : 3 - 9
  • [32] A Security Risk Taxonomy for Prompt-Based Interaction With Large Language Models
    Derner, Erik
    Batistic, Kristina
    Zahalka, Jan
    Babuska, Robert
    IEEE ACCESS, 2024, 12 : 126176 - 126187
  • [33] TrojLLM: A Black-box Trojan Prompt Attack on Large Language Models
    Xue, Jiaqi
    Zheng, Mengxin
    Hua, Ting
    Shen, Yilin
    Liu, Yepeng
    Boloni, Ladislau
    Lou, Qian
    ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 36 (NEURIPS 2023), 2023,
  • [34] Prompt engineering on leveraging large language models in generating response to InBasket messages
    Yan, Sherry
    Knapp, Wendi
    Leong, Andrew
    Kadkhodazadeh, Sarira
    Das, Souvik
    Jones, Veena G.
    Clark, Robert
    Grattendick, David
    Chen, Kevin
    Hladik, Lisa
    Fagan, Lawrence
    Chan, Albert
    JOURNAL OF THE AMERICAN MEDICAL INFORMATICS ASSOCIATION, 2024, 31 (10) : 2263 - 2270
  • [35] Optimizing Large Language Models: A Deep Dive into Effective Prompt Engineering Techniques
    Son, Minjun
    Won, Yun-Jae
    Lee, Sungjin
    APPLIED SCIENCES-BASEL, 2025, 15 (03):
  • [36] Interpretable Online Log Analysis Using Large Language Models with Prompt Strategies
    Liu, Yilun
    Tao, Shimin
    Meng, Weibin
    Wang, Jingyu
    Ma, Wenbing
    Chen, Yuhang
    Zhao, Yanqing
    Yang, Hao
    Jiang, Yanfei
    PROCEEDINGS 2024 32ND IEEE/ACM INTERNATIONAL CONFERENCE ON PROGRAM COMPREHENSION, ICPC 2024, 2024, : 35 - 46
  • [37] Flocks of Stochastic Parrots: Differentially Private Prompt Learning for Large Language Models
    Duan, Haonan
    Dziedzic, Adam
    Papernot, Nicolas
    Boenisch, Franziska
    ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 36 (NEURIPS 2023), 2023,
  • [38] DrugReAlign: a multisource prompt framework for drug repurposing based on large language models
    Wei, Jinhang
    Zhuo, Linlin
    Fu, Xiangzheng
    Zeng, Xiangxiang
    Wang, Li
    Zou, Quan
    Cao, Dongsheng
    BMC BIOLOGY, 2024, 22 (01)
  • [39] Towards Taming Large Language Models with Prompt Templates for Legal GRL Modeling
    de Kinderen, Sybren
    Winter, Karolin
    ENTERPRISE, BUSINESS-PROCESS AND INFORMATION SYSTEMS MODELING, BPMDS 2024, EMMSAD 2024, 2024, 511 : 213 - 228
  • [40] Biomedical knowledge graph-optimized prompt generation for large language models
    Soman, Karthik
    Rose, Peter W.
    Morris, John H.
    Akbas, Rabia E.
    Smith, Brett
    Peetoom, Braian
    Villouta-Reyes, Catalina
    Cerono, Gabriel
    Shi, Yongmei
    Rizk-Jackson, Angela
    Israni, Sharat
    Nelson, Charlotte A.
    Huang, Sui
    Baranzini, Sergio E.
    BIOINFORMATICS, 2024, 40 (09)