Hide and Seek: HIDE Concepts to SEEK explanations for black box models
Presents new explainable AI techniques for understanding vision model predictions, directly relevant to ML research on model interpretability.
AI Summary
Researchers propose HIDE and SEEK (HnS), a novel method for interpreting black-box vision models using concept-based attribution without requiring model weights or labeled concept datasets.
Excerpt
Interpreting machine learning model predictions is a central challenge, particularly for black-box vision models. Concept-based attribution methods interpret models using human-understandable concepts. Existing concept-based methods can only interpret classification and regression models. They require access to model weights or large concept-annotated datasets, and most methods cannot provide image-level attribution. We propose Hide and Seek (HnS), a new method for concept-level attribution that