SoftSAE: Dynamic Top-K Selection for Adaptive Sparse Autoencoders

• URL: http://arxiv.org/abs/2605.06610v2
• Date: Fri, 08 May 2026 10:28:18 GMT
• Title: SoftSAE: Dynamic Top-K Selection for Adaptive Sparse Autoencoders
• Abstract summary: We propose SoftSAE, a sparse autoencoder with a Dynamic Top-K selection mechanism.<n>Our method uses a differentiable Soft Top-K operator to learn an input-dependent sparsity level k.<n> Experimental results confirm that SoftSAE not only finds meaningful features, but also selects the right number of features for each concept.
• Score: 11.728621223411475
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Sparse Autoencoders (SAEs) have become an important tool in mechanistic interpretability, helping to analyze internal representations in both Large Language Models (LLMs) and Vision Transformers (ViTs). By decomposing polysemantic activations into sparse sets of monosemantic features, SAEs aim to translate neural network computations into human-understandable concepts. However, common architectures such as TopK SAEs rely on a fixed sparsity level. They enforce the same number of active features (K) across all inputs, ignoring the varying complexity of real-world data. Natural data often lies on manifolds with varying local intrinsic dimensionality, meaning the number of relevant factors can change significantly across samples. This suggests that a fixed sparsity level is not optimal. Simple inputs may require only a few features, while more complex ones need more expressive representations. Using a constant K can therefore introduce noise in simple cases or miss important structure in more complex ones. To address this issue, we propose SoftSAE, a sparse autoencoder with a Dynamic Top-K selection mechanism. Our method uses a differentiable Soft Top-K operator to learn an input-dependent sparsity level k. This allows the model to adjust the number of active features based on the complexity of each input. As a result, the representation better matches the structure of the data, and the explanation length reflects the amount of information in the input. Experimental results confirm that SoftSAE not only finds meaningful features, but also selects the right number of features for each concept. The source code is available at: https://github.com/St0pien/SoftSAE.

Related papers

• Interpretability Without Tradeoffs: Disentangling Polysemanticity At Equal Predictive Performance [58.49623467782041]
  ELUDe is a method for improving the interpretability of deep neural networks.<n>It breaks latent representations into clear, inspectable sub-units that behave like interpretable features.<n>It requires no explicit training, no labels, and can be applied to pretrained models.
  arXiv  Detail & Related papers  (2026-05-29T13:39:09Z)
• Step-Level Sparse Autoencoder for Reasoning Process Interpretation [48.99201531966593]
  Large Language Models (LLMs) have achieved strong complex reasoning capabilities through Chain-of-Thought (CoT) reasoning.<n>We propose step-level sparse autoencoder (SSAE), which serves as an analytical tool to disentangle different aspects of LLMs' reasoning steps into sparse features.<n> Experiments on multiple base models and reasoning tasks show the effectiveness of the extracted features.
  arXiv  Detail & Related papers  (2026-03-03T14:25:02Z)
• AdaptiveK Sparse Autoencoders: Dynamic Sparsity Allocation for Interpretable LLM Representations [28.447024168930984]
  We propose AdaptiveK SAE (Adaptive Top K Sparse Autoencoders), a novel framework that dynamically adjusts sparsity levels based on the semantic complexity of each input.<n>We show that this complexity-driven adaptation significantly outperforms fixed-sparsity approaches on reconstruction fidelity, explained variance, cosine similarity and interpretability metrics.
  arXiv  Detail & Related papers  (2025-08-24T12:00:41Z)
• Kolmogorov GAM Networks are all you need! [0.6906005491572398]
  Kolmogorov GAM networks are shown to be an efficient architecture for training and inference.<n>They are an additive model with an embedding that is independent of the function of interest.
  arXiv  Detail & Related papers  (2025-01-01T02:46:00Z)
• Interpretability at Scale: Identifying Causal Mechanisms in Alpaca [62.65877150123775]
  We use Boundless DAS to efficiently search for interpretable causal structure in large language models while they follow instructions. Our findings mark a first step toward faithfully understanding the inner-workings of our ever-growing and most widely deployed language models.
  arXiv  Detail & Related papers  (2023-05-15T17:15:40Z)
• Factorizers for Distributed Sparse Block Codes [45.29870215671697]
  We propose a fast and highly accurate method for factorizing distributed block codes (SBCs) Our iterative factorizer introduces a threshold-based nonlinear activation, conditional random sampling, and an $ell_infty$-based similarity metric. We demonstrate the feasibility of our method on four deep CNN architectures over CIFAR-100, ImageNet-1K, and RAVEN datasets.
  arXiv  Detail & Related papers  (2023-03-24T12:31:48Z)
• Identifiable Variational Autoencoders via Sparse Decoding [37.30831737046145]
  We develop the Sparse VAE, a deep generative model for unsupervised representation learning on high-dimensional data. We first show that the Sparse VAE is identifiable: given data drawn from the model, there exists a uniquely optimal set of factors. We empirically study the Sparse VAE with both simulated and real data.
  arXiv  Detail & Related papers  (2021-10-20T22:11:33Z)
• Inducing Transformer's Compositional Generalization Ability via Auxiliary Sequence Prediction Tasks [86.10875837475783]
  Systematic compositionality is an essential mechanism in human language, allowing the recombination of known parts to create novel expressions. Existing neural models have been shown to lack this basic ability in learning symbolic structures. We propose two auxiliary sequence prediction tasks that track the progress of function and argument semantics.
  arXiv  Detail & Related papers  (2021-09-30T16:41:19Z)
• Statistically Meaningful Approximation: a Case Study on Approximating Turing Machines with Transformers [50.85524803885483]
  This work proposes a formal definition of statistically meaningful (SM) approximation which requires the approximating network to exhibit good statistical learnability. We study SM approximation for two function classes: circuits and Turing machines.
  arXiv  Detail & Related papers  (2021-07-28T04:28:55Z)
• A New Modal Autoencoder for Functionally Independent Feature Extraction [6.690183908967779]
  A new modal autoencoder (MAE) is proposed by othogonalising the columns of the readout weight matrix. The results were validated on the MNIST variations and USPS classification benchmark suite. The new MAE introduces a very simple training principle for autoencoders and could be promising for the pre-training of deep neural networks.
  arXiv  Detail & Related papers  (2020-06-25T13:25:10Z)
• Anchor & Transform: Learning Sparse Embeddings for Large Vocabularies [60.285091454321055]
  We design a simple and efficient embedding algorithm that learns a small set of anchor embeddings and a sparse transformation matrix. On text classification, language modeling, and movie recommendation benchmarks, we show that ANT is particularly suitable for large vocabulary sizes.
  arXiv  Detail & Related papers  (2020-03-18T13:07:51Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.