Expander SAEs apply left-d-regular expander masks to TopK SAEs, learning only dn decoder parameters instead of mn and tracing a storage-fidelity frontier that reaches 293x compression with 84% retained performance on Qwen2.5-3B.
Efficient dictionary learning with switch sparse autoencoders
5 Pith papers cite this work. Polarity classification is still indexing.
5
Pith papers citing it
years
2026 5representative citing papers
Sparse autoencoders enable phase synchronization in frozen graph CFD surrogates through Hilbert-identified oscillatory features and SVD-based time-varying rotations.
The survey organizes mechanistic interpretability techniques into a Locate-Steer-Improve framework to enable actionable improvements in LLM alignment, capability, and efficiency.
Topic-aware augmentation makes psychosocial risk factors such as immigration, family issues, and financial crisis more distinct and coherent in the internal representations of suicide ideation detection models.
citing papers explorer
-
Expander Sparse Autoencoders: Parameter-Efficient Dictionaries for Mechanistic Interpretability
Expander SAEs apply left-d-regular expander masks to TopK SAEs, learning only dn decoder parameters instead of mn and tracing a storage-fidelity frontier that reaches 293x compression with 84% retained performance on Qwen2.5-3B.
-
Sparse Autoencoders as a Steering Basis for Phase Synchronization in Graph-Based CFD Surrogates
Sparse autoencoders enable phase synchronization in frozen graph CFD surrogates through Hilbert-identified oscillatory features and SVD-based time-varying rotations.
-
Locate, Steer, and Improve: A Practical Survey of Actionable Mechanistic Interpretability in Large Language Models
The survey organizes mechanistic interpretability techniques into a Locate-Steer-Improve framework to enable actionable improvements in LLM alignment, capability, and efficiency.
-
Beyond Accuracy: Interpreting Topic Representation in Suicide Ideation Detection Models
Topic-aware augmentation makes psychosocial risk factors such as immigration, family issues, and financial crisis more distinct and coherent in the internal representations of suicide ideation detection models.
- Sign-Aware Gated Sparse Autoencoders: Modeling Anticorrelated Features with Bi-Jump-ReLU Activations