Presenting at NAACL 2024 🇲🇽

Jun 15, 2024

Time	Location	Activity
June 17, 16:00-17:30	Ethics, Bias, and Fairness 2, DON ALBERTO 1	Oral presentation of the main conf paper: Instructional Fingerprinting of Large Language Models. We present a pilot study on LLM fingerprinting as a form of very lightweight instruction tuning.
June 18, 9:00-10:30	In-Person Poster Session 4, DON DIEGO 2, 3 & 4	Poster presentation of the main conf paper Mitigating Bias for Question Answering Models by Tracking Bias Influence. We propose to detect the bias level of QA instances by observing its downstream influence in the in-context learning setting.
June 18, 9:00-10:30	In-Person Poster Session 4, DON DIEGO 2, 3 & 4	Poster presentation of the main conf paper: Instructions as Backdoors: Backdoor Vulnerabilities of Instruction Tuning for Large Language Models. We demonstrate that an attacker can inject backdoors by issuing very few malicious instructions and control model behavior through data poisoning.

Presenting Papers

Mitigating Bias for Question Answering Models by Tracking Bias Influence

Mingyu Derek Ma, Jiun-Yu Kao, Arpit Gupta, Yu-Hsiang Lin, Wenbo Zhao, Tagyoung Chung, Wei Wang, Kai-Wei Chang, Nanyun Peng

NAACL, 2024

We propose BMBI, an approach to mitigate the bias of multiple-choice QA models. Based on the intuition that a model would lean to be more biased if it learns from a biased example, we measure the bias level of a query instance by observing its influence on another instance. We then use the bias level detected as an optimization objective to form a multi-task learning setting in addition to the original QA task.

PDF Cite Poster DOI

Instructions as Backdoors: Backdoor Vulnerabilities of Instruction Tuning for Large Language Models

Jiashu Xu, Mingyu Derek Ma, Fei Wang, Chaowei Xiao, Muhao Chen

NAACL, 2024

Our studies demonstrate that an attacker can inject backdoors by issuing very few malicious instructions among thousands of gathered data and control model behavior through data poisoning. Through such instruction attacks, the attacker can achieve over 90% attack success rate across four commonly used NLP datasets, and cause persistent backdoors that are easily transferred to 15 diverse datasets zero-shot.

Instructional Fingerprinting of Large Language Models

Jiashu Xu, Fei Wang^*, Mingyu Derek Ma^*, Pang Wei Koh, Chaowei Xiao, Muhao Chen

NAACL, 2024

We present a pilot study on LLM fingerprinting as a form of very lightweight instruction tuning. Model publisher specifies a confidential private key and implants it as an instruction backdoor that causes the LLM to generate specific text when the key is present. Results on 11 popularly-used LLMs showed that this approach prevents publisher overclaim, maintains robustness against fingerprint guessing and parameter-efficient training, and supports multi-stage fingerprinting akin to MIT License.

Project PDF Cite Code DOI

Last updated on Jun 15, 2024

Conference NAACL

Presenting at AAAI 2024 🇨🇦 Feb 22, 2024 →