Graph Adversarial Learning for the Generation and Detection of Cyber Attacks in Transportation
Ensuring the security of released large language models (LLMs) poses a significant dilemma, as existing mechanisms either compromise ownership rights or raise data privacy concerns. To address this dilemma, the research team introduces TaylorMLP to protect the ownership of released LLMs and prevent their abuse. Specifically, TaylorMLP preserves the ownership of LLMs by transforming the weights of LLMs into parameters of the Taylor series. Instead of releasing the original weights, developers can release the Taylor-series parameters with users, thereby ensuring the security of LLMs. Moreover, TaylorMLP can prevent the abuse of LLMs by adjusting the generation speed. It can induce low-speed token generation for the protected LLMs by increasing the terms in the Taylor series. This intentional delay helps LLM developers prevent potentially large-scale unauthorized uses of their models. Empirical experiments across five datasets and three LLM architectures demonstrate that TaylorMLP induces over 4x increase in latency, producing the tokens precisely matched with original LLMs. Subsequent defensive experiments further confirm that TaylorMLP effectively prevents users from reconstructing the weight values based on downstream datasets.
Language
- English
Project
- Status: Active
- Funding: $60,000.00
-
Contract Numbers:
69A3552348332
-
Sponsor Organizations:
Office of the Assistant Secretary for Research and Technology
University Transportation Centers Program
Department of Transportation
Washington, DC United States 20590 -
Managing Organizations:
Transportation Cybersecurity Center for Advanced Research and Education (CYBER-CARE)
University of Houston
Houston, TX United States -
Performing Organizations:
University of Hawaii, Manoa
2540 Dole Street
Honolulu, HI United States 96822 Houston, Texas United States -
Principal Investigators:
Hu, Xia
- Start Date: 20230701
- Expected Completion Date: 20251231
- Actual Completion Date: 0
- USDOT Program: University Transportation Centers Program
Subject/Index Terms
- TRT Terms: Computer security; Data analysis; Information processing
- Subject Areas: Data and Information Technology; Security and Emergencies; Transportation (General);
Filing Info
- Accession Number: 01953916
- Record Type: Research project
- Source Agency: Transportation Cybersecurity Center for Advanced Research and Education (CYBER-CARE)
- Contract Numbers: 69A3552348332
- Files: UTC, RIP, STATEDOT
- Created Date: Apr 29 2025 4:41PM