arxiv
PublishedJune 30, 2026 at 4:00 AM
—neutral
C$^{2}$R: Cross-sample Consistency Regularization Mitigates Feature Splitting and Absorption in Sparse Autoencoders
Publisher summary· verbatim
arXiv:2606.30609v1 Announce Type: cross Abstract: Sparse Autoencoders (SAEs) are widely used to interpret large language models by decomposing activations into sparse, human-understandable features, but scaling to large dictionaries exposes fundamental challenges. Systematic studies reveal pervasive
Stay posted· Newsletter
A 5-min weekly brief — top movers, price watch, story of the week.
Discussion
No replies yet. Be first.
Related coverage
More from ARXIV
arxivTransformer-Based Inverse Microrheology for Experimental Mechanics at Ultra-High Strain Rates23harxivDeployment Risk Assessment Using Diff-Aware Features: A Case Study at Prime Video23harxivSolarChain-Eval: A Physics-Constrained Benchmark for Trustworthy Economic Agents in Decentralized Energy Markets23harxivForget Narrowly, Retain Broadly: Unlearning as an Asymmetric Generalization Problem23hThe Bubble Brief
WEEKLYRead AI insights every Tuesday — top movers, new releases, story of the week.
Originally published on arxiv ↗