A Deeper Analysis of Block-Sparse Featurizers

A new analysis of the block-sparse featurizer (BSF), introduced by Fel et al. in 2026, finds it still suffers from classic sparse autoencoder failure modes like feature splitting and composition, despite its design for low-dimensional manifold features common in vision. The authors propose architectural changes, including a Tournament Top-K selection rule that significantly reduces feature splitting, and extend the block paradigm to the crosscoder.

Computer Science Machine Learning Submitted on 27 Aug 2026 Title:A Deeper Analysis of Block-Sparse Featurizers View PDF /pdf/2608.27515 HTML experimental https://arxiv.org/html/2608.27515v1 Abstract:The recently introduced block-sparse featurizer BSF; Fel et al., 2026 is similar to a sparse autoencoder SAE , but its atomic unit is a small subspace a block of directions rather than a single direction. It is designed for features that live on low-dimensional manifolds, which are especially frequent in vision. This work studies the BSF's strengths and weaknesses, finding how it still somewhat suffers from classic SAE failure modes, like feature splitting and composition. We propose several architectural changes to the BSF, including a Tournament Top-K selection rule that significantly reduces feature splitting, and we also extend the block paradigm to the crosscoder. Submission history From: Alexandru-Iulius Jerpelea view email /show-email/5f30d5d2/2608.27515 v1 Thu, 27 Aug 2026 09:50:20 UTC 1,239 KB References & Citations Loading... Bibliographic and Citation Tools Bibliographic Explorer What is the Explorer? https://info.arxiv.org/labs/showcase.html arxiv-bibliographic-explorer Connected Papers What is Connected Papers? https://www.connectedpapers.com/about Litmaps What is Litmaps? https://www.litmaps.co/ scite Smart Citations What are Smart Citations? https://www.scite.ai/ Code, Data and Media Associated with this Article alphaXiv What is alphaXiv? https://alphaxiv.org/ CatalyzeX Code Finder for Papers What is CatalyzeX? https://www.catalyzex.com DagsHub What is DagsHub? https://dagshub.com/ Gotit.pub What is GotitPub? http://gotit.pub/faq Hugging Face What is Huggingface? https://huggingface.co/huggingface ScienceCast What is ScienceCast? https://sciencecast.org/welcome Demos Recommenders and Search Tools Influence Flower What are Influence Flowers? https://influencemap.cmlab.dev/ CORE Recommender What is CORE? https://core.ac.uk/services/recommender IArxiv Recommender What is IArxiv? https://iarxiv.org/about 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 https://info.arxiv.org/labs/index.html .