Search papers, labs, and topics across Lattice.
This paper introduces BadLLM-TG, a novel backdoor defense mechanism for NLP models that leverages LLMs to generate effective trigger inversions. The approach uses prompt-driven reinforcement learning, with the victim model's feedback loss as a reward signal, to optimize the trigger generator. The generated triggers are then used in adversarial training to mitigate the backdoor effect, achieving a 76.2% reduction in attack success rate.
LLMs can be prompted to generate effective trigger inversions for backdoor defense, outperforming existing methods by a significant margin.
Backdoor attacks compromise model reliability by using triggers to manipulate outputs. Trigger inversion can accurately locate these triggers via a generator and is therefore critical for backdoor defense. However, the discrete nature of text prevents existing noise-based trigger generator from being applied to nature language processing (NLP). To overcome the limitations, we employ the rich knowledge embedded in large language models (LLMs) and propose a Backdoor defender powered by LLM Trigger Generator, termed BadLLM-TG. It is optimized through prompt-driven reinforcement learning, using the victim model's feedback loss as the reward signal. The generated triggers are then employed to mitigate the backdoor via adversarial training. Experiments show that our method reduces the attack success rate by 76.2\% on average, outperforming the second-best defender by 13.7.
