SwordBench: Evaluating Orthogonality of Steering Image Representations
SwordBench is a new benchmark for evaluating steering of image representations in vision models. It introduces cross-concept robustness and collateral damage metrics to measure the effects of orthogonalization. Experiments show linear SVM achieves good separability and orthogonality but fails to achieve zero collateral damage, while sparse autoencoders perform better. In simpler regimes, standard baselines and optimization-based methods fail to achieve perfect steering. Code will be released on GitHub.
[2605.16372] SwordBench: Evaluating Orthogonality of Steering Image Representations
[Submitted on 10 May 2026]
Title:SwordBench: Evaluating Orthogonality of Steering Image Representations
View a PDF of the paper titled SwordBench: Evaluating Orthogonality of Steering Image Representations, by Vladimir Zaigrajew and 3 other authors
View PDF HTML (experimental)
Abstract:Steering or intervening on model representations at inference time to correct predictions is essential for AI interpretability and safety, yet existing evaluation protocols are limited to ambiguous language modeling tasks. To address this gap, we introduce SwordBench, a benchmark for steering image representations of vision models across multiple backbones and concept removal tasks. Beyond a unified benchmarking suite, we propose new evaluation notions that uncover the second-order effects of orthogonalization among concept activation vectors for pragmatic steering. Specifically, cross-concept robustness measures the stability of concept detection performance across inputs orthogonalized against alternative concepts, and collateral damage quantifies whether steering inadvertently affects model performance on a downstream task for inputs lacking the bias. We find that although a linear support vector machine exhibits superior separability and orthogonality, it fails to achieve zero collateral damage, often trailing sparse autoencoders. In simpler regimes, both standard baselines and optimization-based methods fail to achieve perfect steering. The source code will be made available soon on GitHub.
Subjects:
Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
Cite as: arXiv:2605.16372 [cs.CV]
(or arXiv:2605.16372v1 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2605.16372
arXiv-issued DOI via DataCite (pending registration)
Submission history
From: Vladimir Zaigrajew [view email] [v1] Sun, 10 May 2026 14:45:52 UTC (7,860 KB)
Full-text links:
Access Paper:
View a PDF of the paper titled SwordBench: Evaluating Orthogonality of Steering Image Representations, by Vladimir Zaigrajew and 3 other authors
View PDF
HTML (experimental)
TeX Source
view license
Current browse context:
cs.CV
new | recent | 2026-05
Change to browse by:
cs cs.AI cs.LG
References & Citations
NASA ADS
Google Scholar
Semantic Scholar
Loading...
Data provided by:
Bibliographic Tools
Bibliographic and Citation Tools
Bibliographic Explorer Toggle
Bibliographic Explorer (What is the Explorer?)
Connected Papers Toggle
Connected Papers (What is Connected Papers?)
Litmaps Toggle
Litmaps (What is Litmaps?)
scite.ai Toggle
scite Smart Citations (What are Smart Citations?)
Code, Data, Media
Code, Data and Media Associated with this Article
alphaXiv Toggle
alphaXiv (What is alphaXiv?)
Links to Code Toggle
CatalyzeX Code Finder for Papers (What is CatalyzeX?)
DagsHub Toggle
DagsHub (What is DagsHub?)
GotitPub Toggle
Gotit.pub (What is GotitPub?)
Huggingface Toggle
Hugging Face (What is Huggingface?)
ScienceCast Toggle
ScienceCast (What is ScienceCast?)
Demos
Demos
Replicate Toggle
Replicate (What is Replicate?)
Spaces Toggle
Hugging Face Spaces (What is Spaces?)
Spaces Toggle
TXYZ.AI (What is TXYZ.AI?)
Related Papers
Recommenders and Search Tools
Link to Influence Flower
Influence Flower (What are Influence Flowers?)
Core recommender toggle
CORE Recommender (What is CORE?)
Author
Venue
Institution
Topic
About arXivLabs
arXivLabs: experimental projects with community collaborators
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.
Which authors of this paper are endorsers? | Disable MathJax (What is MathJax?)