Dynamic Adversarial Fine-Tuning Reorganizes Refusal Geometry · 相关技术
相关技术
ODEPLAlanguage model远程代码执行(RCE)研究法辛烷值safetyrsaevaladversarial modelSynthetic Aperture RadarStraight-Through EstimatorSpatial Pivot-Aligned Coordinate-free Embedding (SPACE)SPASupervised Fine-TuningSFTSAFERepresentation AnchoringReferring expression comprehension (REC)PromptGEOGANEffort Metric AttentionDiTDAPTARGANNCLEAR