Skip to content
AI News HubLIVE
Original source2 min read

FailSAE: Towards Interpretable Failure Prediction for Vision-Language Models via Sparse Autoencoders

Summary

FailSAE uses sparse autoencoders (SAEs) to predict failures of vision-language models while preserving interpretability. The authors frame failure prediction as classification over sparse latent activations and apply a three-stage failure-aware training pipeline. Experiments outperform baselines; concept-level analysis reveals failures correspond to a shift away from class-specific concepts toward ambiguous or style-related concepts, and the learned latent directions also support runtime failure recovery.

SourcearXiv Computer VisionAuthor: Jie Ma, Zongxi Liu, Yi Zhu
FailSAE: Towards Interpretable Failure Prediction for Vision-Language Models via Sparse Autoencoders
Report an error

The correction channel is not available yet. You can copy the article reference below for later.

Correction instructions
Read article

[Submitted on 2 Sep 2026]

Title:FailSAE: Towards Interpretable Failure Prediction for Vision-Language Models via Sparse Autoencoders

View a PDF of the paper titled FailSAE: Towards Interpretable Failure Prediction for Vision-Language Models via Sparse Autoencoders, by Jie Ma and 2 other authors

View PDF HTML (experimental)

Abstract:Vision-language models (VLMs), such as CLIP, have achieved strong performance across multimodal tasks by aligning visual and textual representations in a shared embedding space. As VLMs are increasingly used for high-stakes domains, failure prediction becomes critical for risk-aware deployment and human intervention. Existing failure prediction methods typically rely on confidence scores or auxiliary classifiers. Although these methods are effective on predicting VLM failures, they provide limited interpretability. In this work, we investigate the use of Sparse Autoencoders (SAEs) for interpretable failure prediction in VLMs. We formulate failure prediction as a classification task over sparse SAE latent activations and introduce a three-stage failure-aware training pipeline that encourages the learned latent directions to remain interpretable while becoming more informative for failure prediction. Our experiments show that the resulting framework outperforms the evaluated baselines in failure prediction. Further analysis suggests that failure-aware training encourages SAE latent directions to capture more class-specific concepts. We also use the SAE to provide a concept-level analysis of how model representations change during failures, revealing a shift from class-specific concepts toward more ambiguous or style-related concepts. Finally, we explore how the learned SAE latent directions can support runtime failure recovery.

Subjects:

Computer Vision and Pattern Recognition (cs.CV)

ACM classes: I.4.7

Cite as: arXiv:2609.04276 [cs.CV]

(or arXiv:2609.04276v1 [cs.CV] for this version)

https://doi.org/10.48550/arXiv.2609.04276

arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Yi Zhu [view email] [v1] Wed, 2 Sep 2026 23:52:47 UTC (1,547 KB)

Full-text links:

Access Paper:

View a PDF of the paper titled FailSAE: Towards Interpretable Failure Prediction for Vision-Language Models via Sparse Autoencoders, by Jie Ma and 2 other authors

View PDF

HTML (experimental)

TeX Source

view license

Current browse context:

cs.CV

new | recent | 2026-09

Change to browse by:

cs

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?)

Key points and analysis

Article intelligence

EngineersAdvanced

Key points

  • Proposes FailSAE, a sparse-autoencoder-based approach for interpretable failure prediction in vision-language models.
  • Formulates failure prediction as classification over sparse SAE latent activations, with a three-stage failure-aware training pipeline.
  • Outperforms evaluated baselines and shows failure-aware training captures more class-specific concepts.
  • Provides concept-level insights linking failures to a shift from class-specific toward ambiguous/style-related concepts, and discusses runtime failure recovery.

Highlights and analysis are generated automatically and may contain errors. Check the original source.

More recent stories on this topic · Models