{"ID":2865136,"CreatedAt":"2026-06-01T04:54:23.091178241Z","UpdatedAt":"2026-06-01T04:54:23.091178241Z","DeletedAt":null,"paper_url":"https://arxiv.org/abs/2509.22015","arxiv_id":"2509.22015","title":"Concept-SAE: Active Causal Probing of Visual Model Behavior","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.","short_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 Conc...","url_abs":"https://arxiv.org/abs/2509.22015","url_pdf":"https://arxiv.org/pdf/2509.22015v1","authors":"[\"Jianrong Ding\",\"Muxi Chen\",\"Chenchen Zhao\",\"Qiang Xu\"]","published":"2025-09-26T07:51:03Z","proceeding":"cs.LG","tasks":"[\"cs.LG\"]","methods":"[]","has_code":false}