paper-with-me

홈 › Papers

Interpreting vision transformers via residual replacement model

2025-09-22 · Jinyeong Kim, Junhyeok Kim, Yumin Shim, Joohyeok Kim, Sunyoung Jung, Seong Jae Hwang arxiv

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.

📄 PDF Abstract BibTeX arXiv:2509.17401

Code (0)

등록된 구현이 없습니다.

Similar Papers 제목 키워드 기반

3D Vision with Transformers: A Survey

2022-08-08 · Jean Lahoud, Jiale Cao, Fahad Shahbaz Khan, Hisham Cholakkal 외

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 EstimationSurvey

From Layers to Submodules: Rethinking Granularity in Replacement-Based LLM Compression

2026-06-01 · Elia Cunegatti, Marcus Vukojevic, Erik Nielsen, Giovanni Iacca arxiv

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

2026-05-07 · Pengqi Lu arxiv

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

2024-03-28 · CVPR 2024 1 · Wei Dong, Xing Zhang, Bihui Chen, Dawei Yan 외

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-tuning

Zero-Ablation Overstates Register Content Dependence in DINO Vision Transformers

2026-04-15 · Felipe Parodi, Jordan Matelsky, Melanie Segado arxiv

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…