Inside the Visual Mind: Neuroscience-Motivated Concept Circuits for Interpreting and Steering Vision Transformers
Quick Answer
ViSAE enhances the interpretability of Vision Transformers (ViTs) by using neuroscience-inspired concept circuits, achieving a 48.2% improvement in worst-group accuracy on the WaterBirds benchmark.
Quick Take
ViSAE enhances the interpretability of Vision Transformers (ViTs) by using neuroscience-inspired concept circuits, achieving a 48.2% improvement in worst-group accuracy on the WaterBirds benchmark. This toolbox includes a probing suite with 64K images and a 16K concept vocabulary, improving concept coverage by 20x and interpretation accuracy by 28.7%. ViSAE enables better auditing and steering of ViT behavior.
Key Points
- ViSAE improves worst-group accuracy on WaterBirds by 48.2%, outperforming existing methods by 23.8%.
- The probing suite features 64K images and a 16K visually grounded concept vocabulary.
- Concept coverage efficiency is enhanced by 20x compared to ImageNet.
- Top-down and bottom-up algorithms recover ViT inner workings through concept circuits.
- ViSAE addresses challenges in adapting sparse autoencoders for ViT interpretation.
Article Content
From source RSS / original summaryarXiv:2606. 06664v1 Announce Type: new Abstract: Despite high accuracy, Vision Transformer (ViT) predictions can be driven by spurious cues, raising the need to understand their inner workings before safe deployment. Sparse autoencoders (SAEs) provide a promising lens for decomposing model representations into human-interpretable concepts, yet adapting SAE-based interpretation to ViTs remains challenging due to limited control over concept coverage and subjective, non-scalable feature interpretation.
To fill the gaps, motivated by neuroscience-inspired principles, we propose ViSAE, a mechanistic interpretability toolbox for understanding ViT inner workings through concept circuits. ViSAE consists of three components: (1) A probing suite with 64K images and a 16K visually grounded concept vocabulary, improving concept coverage efficiency by 20x over ImageNet and interpretation accuracy by 28. 7% over existing concept sets.
(2) Top-down concept reading and Bottom-up circuit tracing algorithms that automatically recover ViT inner workings via concept circuits. (3) Applications for auditing and steering ViT behavior. Through concept editing, ViSAE improves the worst-group accuracy on WaterBirds by 48. 2%, outperforming existing methods by 23. 8%. Our data and code: https://github. com/deep-real/ViSAE.
Reader Mode unavailable (could not extract clean content).
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.CV
See more →LLM-Guided ANN Index Optimization for Human-Object Interaction Retrieval
A phase-aware LLM agent optimizes human-object interaction retrieval, outperforming Optuna TPE by 33.3% and VDTuner by 34.2% on the HICO-DET benchmark. This method enhances throughput by 15.3x over UniIR and demonstrates strong transferability across vector database management systems.