Zhiyu Zhu

Faithfulness Under the Distribution: A New Look at Attribution Evaluation

Label-Free Attribution for Interpretability

Adversarial Bottleneck Method for Vision-Language Large Model Explainability

LACE: Lightweight Attribution-guided Concept Evolution for Continual Learning

Signed-dictionary and Nonnegative-activation Decomposition for Concept Bottleneck Models

Can we trust the attribution method?

Conceptual Archetype Decomposition for Interpretable and Generalizable Model Decisions

C2MF: Consistent and Concept-Unified Matrix Factorization for Interpretable and Robust Concept Discovery

Splitting & Integrating: Out-of-Distribution Detection via Adversarial Gradient Attribution

Narrowing Information Bottleneck Theory for Multimodal Image-Text Representations Interpretability

Splitting & Integrating: Out-of-Distribution Detection via Adversarial Gradient Attribution

SSLA: A Generalized Attribution Method for Interpreting Self-Supervised Learning without Downstream Task Dependency

K&L: Penetrating Backdoor Defense with Key and Locks

PAR-AdvGAN: Improving Adversarial Attack Capability with Progressive Auto-Regression AdvGAN

Unbiased Attribution with Intrinsic Information

Rethinking Lipschitzness Data-free Backdoor Defense

Feature Level Instance Attribution