Unelicitable Backdoors Via Cryptographic Transformer CircuitsAndis Draguns,Andrew Gritsevskiy,Sumeet Ramesh Motwani,Christian Schroeder de WittNeurIPS 2024(2024)引用 1|浏览15关键词Backdoor attacks,Transformers,handcrafting model parameters,cryptographic circuitsAI 理解论文溯源树样例生成溯源树,研究论文发展脉络Chat Paper正在生成论文摘要