analiza aiqbench
This arXiv paper is squarely about AI interpretability: it applies sparse autoencoders to analyse 6.6K features across the layers of vision transformers and introduces a "residual replacement model" that approximates the original computation with simpler, human-readable circuits. The reported contributions are mechanistic insight into how ViTs build up from low-level patterns to semantics, and a demonstration of debiasing spurious correlations. For readers, the key question is how far the faithfulness and simplification claims hold on larger models and tasks beyond the paper's evaluation, since interpretability methods often look cleaner in controlled settings than in deployment.
Interpreting vision transformers via residual replacement model
arXiv:2509.17401v2 Announce Type: replace-cross Abstract: 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.
Abstract: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.
Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
arXiv:2509.17401 [cs.CV]
(or arXiv:2509.17401v2 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2509.17401 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 217 cuvinte
Subiecte conexe
The Transformer Revolution, Part 1: Dynamic Processing through Output- Weight Interconnections
arXiv · 5 aug. 2026
Align, Unify, Suppress, Route: A Coherentist View of Transformer Computation
arXiv · 25 aug. 2026
Interior interpretability with attention rollout: contraction and propagation profiles in Transformers
arXiv · 28 iul. 2026
Fully Interpretable Minimal Transformers: From Geometry to Algorithm
arXiv · 8 oct. 2026