The Sparse Feature Budget Becomes the Interpretability Dial
Nathanaël Jacquier, Maria Vakalopoulou, and Mahdi S. Hosseini's June 2026 arXiv paper studies a quiet interpretability design choice: whether a Top-k sparse autoencoder should rely only on a hard feature budget, or whether soft sparsity regularizers can make its features more usable for audits.
The Hard Budget
The paper, arXiv:2606.27321 [cs.LG], was submitted on June 25, 2026. arXiv lists the exact title as Beyond the Hard Budget: Sparsity Regularizers for More Interpretable Top-k Sparse Autoencoders, by Nathanaël Jacquier, Maria Vakalopoulou, and Mahdi S. Hosseini. Its subject categories are Machine Learning and Artificial Intelligence.
A sparse autoencoder is an interpretability instrument: it tries to turn dense model representations into a larger set of sparse latent features that humans can inspect. The Top-k variant enforces sparsity by retaining only the k largest latent activations for each input and zeroing the rest. That hard budget is attractive because it avoids some problems of older L1-penalized sparse autoencoders, including shrinkage and dead latents.
The new paper asks whether that architectural lesson was overlearned. If Top-k was designed to avoid a soft L1 penalty, should all soft sparsity pressure disappear? The authors' answer is no. A hard feature budget and a carefully placed soft regularizer can be complementary.
What the Paper Tests
The authors introduce two regularizers that act before Top-k selection. The first is an L1 penalty on off-support units: latents that are not selected for a given sample. The second is a scale-invariant L1/L2 ratio penalty that concentrates the code onto fewer effective units. Both are applied only to batch-active units, meaning units selected at least once in the batch.
The evaluation uses ImageNet-1K and Open Images V7, with embeddings from three frozen vision foundation models: CLIP ViT-L/14, SigLIP2, and a supervised ViT-L/16. The paper tests multiple Top-k budgets, including k values of 32, 64, and 128, and measures reconstruction quality, monosemanticity, class purity, qualitative feature coherence, robustness to inference-time k, and linear probing under activation truncation.
The reported finding is not that sparsity alone solves interpretability. It is narrower and more useful: both regularizers improve monosemanticity and class purity without hurting reconstruction quality across the tested settings. The L1/L2 regularizer also makes reconstruction more robust when the number of retained units at inference differs from the training budget, and it improves small-budget linear probing by putting more discriminative information into the leading activations.
Why the Mask Matters
The active-unit mask is an important safety detail. The paper argues that penalizing all unselected units directly can kill latents because those units receive no reconstruction gradient. The authors test this by removing the mask and report more dead neurons in nearly every configuration, sometimes by one to two orders of magnitude. They identify the mask as necessary for both stability and the interpretability gains.
That is the kind of detail governance reviews often miss. A model card might say "trained sparse autoencoders for interpretability." But a useful audit wants to know how sparsity was imposed, what was penalized, which units could receive gradients, how dead latents were measured, and whether the feature dictionary stayed usable across budgets.
Interpretability depends on training procedure, not only on the existence of a readable artifact. A feature browser built from unstable latents can create a false sense of inspection: lots of labels, plenty of examples, and too little evidence that the dictionary is faithful enough for the question being asked.
Audit Value
This paper belongs beside mechanistic interpretability, but its governance value is not a dramatic circuit discovery. It is a method-design lesson for routine feature audits. The hard k is not a neutral constant. It is an interpretability dial. If a dictionary overfits to a single training budget, auditors may see different stories when they inspect the same representation under a slightly different feature count.
The L1/L2 result is especially relevant because audits often operate under budget pressure. A regulator, safety team, or outside researcher may not inspect hundreds of latents per sample. If discriminative information is front-loaded into fewer leading units without losing full-budget accuracy, small-budget inspections become less arbitrary. That does not make them complete, but it makes the evidence surface more usable.
The Spiralist reading is practical: internal transparency has knobs. A sparse feature is not just discovered; it is produced by an architecture, a loss, a mask, a dataset, a frozen backbone, and a budget. The governance question is whether those knobs are recorded and repeatable.
Limits
The paper studies vision foundation model embeddings, not deployed decision systems or language-model agents. Its claims are about ImageNet-1K, Open Images V7, three frozen vision encoders, and the metrics the authors use. Monosemanticity and class purity are useful proxies, but they are not proof that a feature captures the exact concept a human would use in every setting.
It also does not show that SAE features are complete causal explanations. A sparse autoencoder can improve legibility while still missing interactions, dataset artifacts, rare concepts, or downstream risks. For safety cases, SAE evidence should travel with behavioral evaluations, interventions, ablations, and uncertainty about what the feature dictionary fails to represent.
Method Card
A serious interpretability report using Top-k SAEs should include a method card: source model and layer, dataset, embedding preprocessing, dictionary size, Top-k budget, regularizer type and strength, active-unit mask rule, reconstruction metric, monosemanticity metric, class-purity method, dead-latent count, qualitative inspection protocol, inference-time k sensitivity, and whether the results were stable across seeds.
The claim should be proportionate. "This regularizer improved feature coherence under these conditions" is audit-grade language. "We made the model interpretable" is not. The sparse feature budget becomes useful only when the dial setting is visible.
Sources
- Nathanaël Jacquier, Maria Vakalopoulou, and Mahdi S. Hosseini, Beyond the Hard Budget: Sparsity Regularizers for More Interpretable Top-k Sparse Autoencoders, arXiv:2606.27321 [cs.LG], submitted June 25, 2026.
- arXiv PDF and HTML versions: PDF and experimental HTML, reviewed for authorship, date, Top-k SAE method, regularizers, datasets, models, monosemanticity and reconstruction claims, active-mask ablation, and stated future directions.
- Related pages: Sparse Autoencoders, Mechanistic Interpretability, AI Audits and Assurance, AI Safety Cases, and The Sparse Circuit Becomes the Audit Budget.