Safety
Prototype-Grounded Concept Models for Verifiable Concept Alignment
arXiv:2604.16076v1 Announce Type: cross Abstract: Concept Bottleneck Models (CBMs) aim to improve interpretability in Deep Learning by structuring predictions through human-understandable concepts, bu
arXiv:2604.16076v1 Announce Type: cross Abstract: Concept Bottleneck Models (CBMs) aim to improve interpretability in Deep Learning by structuring predictions through human-understandable concepts, but they provide no way to verify whether learned concepts align with the human's intended meaning, hurting interpretability. We introduce Prototype-Grounded Concept Models (PGCMs), which ground concepts in learned visual prototypes: image parts that serve as explicit evidence for the concepts. This grounding enables direct inspection of concept semantics and supports targeted human intervention at the prototype level to correct misalignments. Empirically, PGCMs match the predictive performance of state-of-the-art CBMs while substantially improving transparency, interpretability, and intervenability.
Related
- PerMix-RLVR: Preserving Persona Expressivity under Verifiable-Reward Alignment
- Rethinking Token-Level Credit Assignment in RLVR: A Polarity-Entropy Analysis
- Calibration Collapse Under Sycophancy Fine-Tuning: How Reward Hacking Breaks Uncertainty Quantification in LLMs
- What's In My Human Feedback? Learning Interpretable Descriptions of Preference Data
Source: arXiv cs.AI | 2026-04-20