When Interpretability Becomes a Liability: Adversarial Attacks on CBM Concept Layers 文章

ArXiv CS.CV2026-05-26NEWSen作者: Aditya Sridhar
查看原文 →
关系图谱

When Interpretability Becomes a Liability: Adversarial Attacks on CBM Concept Layers · 相关技术

相关技术

ODErsamachine learninginterpretabilitydivide-and-conquer partitioningadversarial modelSynthetic Aperture RadarStraight-Through EstimatorStanSpatial Pivot-Aligned Coordinate-free Embedding (SPACE)SPASEMParts-of-Speech (POS) tagsPPRNPUHISForEffort Metric AttentionConcept Bottleneck Models (CBM)BMSARGANNMacTable