大模型对齐与安全

What Your Features Reveal: Data-Efficient Black-Box Feature Inversion Attack for Split DNNs

Zhihan Ren, Lijun He, Jiaxi Liang, Xinzhu Fu, Haixia Bi, Fan Li

机构 * Xi’an Jiaotong University（西安交通大学） 纠错

专题命中 越狱攻击 ：alignment(abstract)