ShadowFPT: Backdooring Federated Prompt Tuning via Shadow Triggers
In brief
- ShadowFPT exploits lightweight prompts in federated learning to inject undetectable backdoors.
- The attack manipulates data representation space using a local 'Shadow Trigger'.
- It maintains clean accuracy while increasing attack success rate from 19.81% to 90.36%.
A new backdoor attack, ShadowFPT, targets vision-language models in federated learning. It exploits lightweight prompt parameters while remaining undetectable in updates sent to the server. The attack manipulates data representation space to induce targeted misclassifications.
This summary gives you the essentials. The full article, its figures and quotes are on cs.AI updates on arXiv.org.
Read the original articleAI-generated summary of the source, checked by rules. Report an error