aiqbench analysis
This research introduces Sign-Aware Gated Sparse Autoencoders (SA-GSAE), a novel architecture that allows single latent variables to represent bidirectional semantic axes, thereby improving parameter efficiency and reducing dead latents in Large Language Models. By integrating two-sided gated sparsity with Bi-Jump-ReLU activations, the method effectively models anticorrelated features without splitting them across multiple dimensions, offering significant gains over standard non-negative SAE variants on mid-depth hookpoints of Pythia-1B and SmolLM3-3B. While full-width implementations face reproducibility challenges regarding reconstruction collapse, the proposed configuration adjustments ensure robust performance for scalable interpretability workflows.
Sign-Aware Gated Sparse Autoencoders: Modeling Anticorrelated Features with Bi-Jump-ReLU Activations
arXiv:2605.28149v2 Announce Type: replace Abstract: Sparse Autoencoders (SAEs) extract interpretable features from Large Language Model activations, but standard variants enforce non-negative latents, so a bidirectional semantic axis (e.g., "pressure too high" vs. "pressure too low") must be split across two latents, wasting dictionary capacity on anticorrelated features. We propose the Sign-Aware Gated SAE (SA-GSAE), which combines two-sided gated sparsity, signed shrinkage-free magnitudes, and auxiliary gate supervision in a new Bi-Jump-ReLU activation, so that a single latent carries both polarities of one decoder direction; parameter accounting shows sign-awareness stays parameter-efficient even when anticorrelated pairs are rare. Across three mid-depth hookpoints on Pythia-1B and SmolLM3-3B (six cells, three seeds), a half-width SA-GSAE empirically dominates the aggregate mean frontier of a full-width Gated SAE on three of six cells, matches its R^2 within 0.025 on the remaining three, and cuts dead fraction by 0.35-0.82 absolute at matched L_0 = 64 on all six. Ablations show the two-sided gate and the auxiliary loss are essential whereas per-polarity asymmetry is not; we recommend the fully tied symmetric variant as the default. A blinded semantic audit finds nameable opposition between a latent's two sides is rare for SA-GSAE and all tested baselines, while sign-conditioned interventions show a single signed latent acts as a bidirectional causal dial where a pair of "opposite" non-negative latents does not; we scope interpretability claims accordingly. At full width, SA-GSAE is over-parameterized and its reported configuration exhibits a reproducible reconstruction collapse at the SmolLM3-3B residual-stream site; the recommended configuration (small threshold initialization with dead-latent threshold resets) prevents it.
Abstract:Sparse Autoencoders (SAEs) extract interpretable features from Large Language Model activations, but standard variants enforce non-negative latents, so a bidirectional semantic axis (e.g., "pressure too high" vs. "pressure too low") must be split across two latents, wasting dictionary capacity on anticorrelated features. We propose the Sign-Aware Gated SAE (SA-GSAE), which combines two-sided gated sparsity, signed shrinkage-free magnitudes, and auxiliary gate supervision in a new Bi-Jump-ReLU activation, so that a single latent carries both polarities of one decoder direction; parameter accounting shows sign-awareness stays parameter-efficient even when anticorrelated pairs are rare. Across three mid-depth hookpoints on Pythia-1B and SmolLM3-3B (six cells, three seeds), a half-width SA-GSAE empirically dominates the aggregate mean frontier of a full-width Gated SAE on three of six cells, matches its R^2 within 0.025 on the remaining three, and cuts dead fraction by 0.35-0.82 absolute at matched L_0 = 64 on all six. Ablations show the two-sided gate and the auxiliary loss are essential whereas per-polarity asymmetry is not; we recommend the fully tied symmetric variant as the default. A blinded semantic audit finds nameable opposition between a latent's two sides is rare for SA-GSAE and all tested baselines, while sign-conditioned interventions show a single signed latent acts as a bidirectional causal dial where a pair of "opposite" non-negative latents does not; we scope interpretability claims accordingly. At full width, SA-GSAE is over-parameterized and its reported configuration exhibits a reproducible reconstruction collapse at the SmolLM3-3B residual-stream site; the recommended configuration (small threshold initialization with dead-latent threshold resets) prevents it.
Machine Learning (cs.LG)
arXiv:2605.28149 [cs.LG]
(or arXiv:2605.28149v2 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2605.28149 Focus to learn more arXiv-issued DOI via DataCite
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.
Full text under the publisher's open licence · 343 words
