English

Concept-SAE: Active Causal Probing of Visual Model Behavior

Machine Learning 2025-09-29 v1

Abstract

Standard Sparse Autoencoders (SAEs) excel at discovering a dictionary of a model's learned features, offering a powerful observational lens. However, the ambiguous and ungrounded nature of these features makes them unreliable instruments for the active, causal probing of model behavior. To solve this, we introduce Concept-SAE, a framework that forges semantically grounded concept tokens through a novel hybrid disentanglement strategy. We first quantitatively demonstrate that our dual-supervision approach produces tokens that are remarkably faithful and spatially localized, outperforming alternative methods in disentanglement. This validated fidelity enables two critical applications: (1) we probe the causal link between internal concepts and predictions via direct intervention, and (2) we probe the model's failure modes by systematically localizing adversarial vulnerabilities to specific layers. Concept-SAE provides a validated blueprint for moving beyond correlational interpretation to the mechanistic, causal probing of model behavior.

Keywords

Cite

@article{arxiv.2509.22015,
  title  = {Concept-SAE: Active Causal Probing of Visual Model Behavior},
  author = {Jianrong Ding and Muxi Chen and Chenchen Zhao and Qiang Xu},
  journal= {arXiv preprint arXiv:2509.22015},
  year   = {2025}
}
R2 v1 2026-07-01T05:58:08.355Z