Interpreting vision transformers via residual replacement model
How do vision transformers (ViTs) represent and process the world? This paper addresses this long-standing question through the first systematic analysis of 6.6K features across all layers, extracted via sparse autoencoders, and by introducing the residual replacement model, which replaces ViT computations with interpretable features in the residual stream. Our analysis reveals not only a feature evolution from low-level patterns to high-level semantics, but also how ViTs encode curves and spatial positions through specialized feature types. The residual replacement model scalably produces a faithful yet parsimonious circuit for human-scale interpretability by significantly simplifying the original computations. As a result, this framework enables intuitive understanding of ViT mechanisms. Finally, we demonstrate the utility of our framework in debiasing spurious correlations.
Code (0)
등록된 구현이 없습니다.
Similar Papers 제목 키워드 기반
3D Vision with Transformers: A Survey
The success of the transformer architecture in natural language processing has recently triggered attention in the computer vision field. The transformer has been used as a replacement for the widely used convolution ope…
Pose EstimationSurveyFrom Layers to Submodules: Rethinking Granularity in Replacement-Based LLM Compression
Post-training compression of Large Language Models (LLMs) removes entire architectural components, either deleting them or replacing them with fitted modules. Existing replacement-based methods share two design constrain…
Mean Mode Screaming: Mean--Variance Split Residuals for 1000-Layer Diffusion Transformers
Scaling Diffusion Transformers (DiTs) to hundreds of layers introduces a structural vulnerability: networks can enter a silent, mean-dominated collapse state that homogenizes token representations and suppresses centered…
Low-Rank Rescaled Vision Transformer Fine-Tuning: A Residual Design Approach
Parameter-efficient fine-tuning for pre-trained Vision Transformers aims to adeptly tailor a model to downstream tasks by learning a minimal set of new adaptation parameters while preserving the frozen majority of pre-tr…
image-classificationImage Classificationparameter-efficient fine-tuningZero-Ablation Overstates Register Content Dependence in DINO Vision Transformers
Zero-ablation -- replacing token activations with zero vectors -- is widely used to probe token function in vision transformers. Register zeroing in DINOv2+registers and DINOv3 produces large drops (up to $-36.6$\,pp cla…