CAFD improves DNN fault detection by 18.3% on average using a novel concept-based feature from VLMs alongside other signals, outperforming five baselines on multiple models and datasets.
Evaluating the robustness of test selection methods for deep neural networks,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
CAFD: Concept-Aware DNN Fault Detection using VLMs
CAFD improves DNN fault detection by 18.3% on average using a novel concept-based feature from VLMs alongside other signals, outperforming five baselines on multiple models and datasets.