Concept-based vision model explanations without model weights or labeled data
Source published ·Modelwire updated
Original coverage: arXiv cs.LG ↗·How Modelwire adds context
The development
Researchers introduce Hide and Seek, a weight-free method for extracting human-interpretable concept-level explanations from black-box vision models without requiring labeled concept datasets. The technique leverages vision-language models to identify and localize concepts within images, then measures their influence through masking. This addresses a critical gap in model interpretability tooling: existing concept attribution methods demand either model internals access or expensive annotation workflows. The approach expands interpretability coverage across diverse vision architectures, lowering barriers for practitioners who need explainability without privileged access. For organizations deploying vision systems under regulatory scrutiny or in high-stakes domains, this represents a practical step toward post-hoc transparency without architectural constraints.
Modelwire’s AI-generated summary of coverage from arXiv cs.LG.
Modelwire analysis
ExplainerOur AI-generated reading of the wider context and the next developments to watch.
The key constraint Hide and Seek removes is the need for labeled concept datasets. Prior work like DisParQ (from two days ago) learns concepts through self-supervised discovery, but still requires model internals or frozen backbones. Hide and Seek works on any frozen vision model as a pure external tool, using a separate vision-language model as the concept oracle.
This sits directly alongside DisParQ and the deception detection work from last week. All three are mechanistic interpretability papers, but they operate at different levels of access. DisParQ patches the foundation model itself; Hide and Seek treats the model as a black box. The deception detection work (Caught in the Act) also uses probes on hidden states, but requires white-box access. Hide and Seek's contribution is lowering the access barrier for concept extraction specifically, which matters for practitioners auditing third-party or proprietary models in regulated domains.
If Hide and Seek's concept attributions correlate with human-labeled ground truth on standard benchmarks (like ImageNet concept subsets) at >0.75 Spearman correlation, the method has real diagnostic value. If correlation drops below 0.60, the vision-language model's concept definitions are too noisy for reliable explanations, and the approach remains a visualization tool rather than a verification tool.
This interpretation is generated from the summary above and the archive coverage cited below. Our methodology · Report an error
Coverage behind this analysis
These archive entries ground the connection in our analysis. They are ordered by source publication date, with links to our coverage and the original sources.
·arXiv cs.LG
Self-supervised vision models learn interpretable part concepts without language
DisParQ advances interpretability in vision models by learning discrete, spatially grounded concepts without language or class labels. The method assigns image patches to learnable prototypes and captures concept variation through continuous residuals, enabling inspection of what frozen self-supervised backbones actually rely on. This addresses a critical gap in foundation model transparency: most concept-based approaches require…
MentionsHide and Seek · vision-language models
How this coverage is produced
Modelwire uses AI to generate summaries and context from source headlines, snippets, and selected archive coverage. Automated checks do not verify every claim, and items are not routinely reviewed by a person before publication. Zacaria Solis operates the site. Read the linked source for the full evidence and report errors through our corrections process.
Modelwire summarizes, we don’t republish. arXiv cs.LG originally reported this story as “Hide and Seek: HIDE Concepts to SEEK explanations for black box models”. The full content lives on arxiv.org. If you’re a publisher and want a different summarization policy for your work, see our takedown page.