Hybrid Adversarial Defence for NLU Tasks

Arxiv CS.CL (NLP/language models)

Proposes a hybrid defence framework that jointly addresses hallucination and adversarial manipulation in LLMs. The approach combines existing defences that typically tackle each problem separately.

llm

Arxiv CS.CL (NLP/language models)·Manar AbouzaidYang WangChenghua LinStuart E. Middleton

7 days ago