Train One Sparse Autoencoder Across Multiple Sparsity Budgets to Preserve Interpretability and Accuracy

• URL: http://arxiv.org/abs/2505.24473v2
• Date: Thu, 05 Jun 2025 11:42:24 GMT
• Title: Train One Sparse Autoencoder Across Multiple Sparsity Budgets to Preserve Interpretability and Accuracy
• Abstract summary: We introduce a novel training objective, emphHierarchicalTopK, which trains a single SAE to optimise reconstructions across multiple sparsity levels simultaneously.<n>Experiments show that HierarchicalTopK preserves high interpretability scores even at higher sparsity.
• Score: 3.2248482136498433
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Sparse Autoencoders (SAEs) have proven to be powerful tools for interpreting neural networks by decomposing hidden representations into disentangled, interpretable features via sparsity constraints. However, conventional SAEs are constrained by the fixed sparsity level chosen during training; meeting different sparsity requirements therefore demands separate models and increases the computational footprint during both training and evaluation. We introduce a novel training objective, \emph{HierarchicalTopK}, which trains a single SAE to optimise reconstructions across multiple sparsity levels simultaneously. Experiments with Gemma-2 2B demonstrate that our approach achieves Pareto-optimal trade-offs between sparsity and explained variance, outperforming traditional SAEs trained at individual sparsity levels. Further analysis shows that HierarchicalTopK preserves high interpretability scores even at higher sparsity. The proposed objective thus closes an important gap between flexibility and interpretability in SAE design.

Related papers

• The Mechanism of Weak-to-Strong Generalization: Feature Elicitation from Latent Knowledge [51.56484100374058]
  Weak-to-strong (W2S) generalization is proposed as an approach to aligning AI systems.<n>We study W2S in the setting of reward-model learning with two-layer neural networks.<n>We prove that the strong model efficiently learns task $$, eliciting its pre-trained knowledge while retaining general capabilities.
  arXiv  Detail & Related papers  (2026-05-13T02:35:58Z)
• Improving Robustness In Sparse Autoencoders via Masked Regularization [21.265992727881514]
  We propose a masking-based regularization that randomly replaces tokens during training to disrupt co-occurrence patterns.<n>This improves robustness across SAE architectures and sparsity levels reducing absorption, enhancing probing performance, and narrowing the OOD gap.
  arXiv  Detail & Related papers  (2026-04-07T21:56:23Z)
• Latent-Space Contrastive Reinforcement Learning for Stable and Efficient LLM Reasoning [16.244366307890832]
  We propose textbfDeepLatent Reasoning (DLR), a latent-space bidirectional contrastive reinforcement learning framework.<n>This framework shifts the trial-and-error cost from expensive token-level full sequence generation to the continuous latent manifold.<n> Experiments demonstrate that DLR achieves more stable training convergence, supports longer-horizon reasoning chains, and facilitates the sustainable accumulation of reasoning capabilities.
  arXiv  Detail & Related papers  (2026-01-24T03:18:22Z)
• Data Whitening Improves Sparse Autoencoder Learning [0.0]
  We show that applying PCA Whitening to input activations improves SAE performance across multiple metrics.<n>We evaluate both ReLU and Top-K SAEs across diverse model architectures, widths, and sparsity regimes.
  arXiv  Detail & Related papers  (2025-11-17T23:20:58Z)
• Beyond Redundancy: Diverse and Specialized Multi-Expert Sparse Autoencoder [59.89996751196727]
  Sparse autoencoders (SAEs) have emerged as a powerful tool for interpreting large language models.<n>SAEs' hidden layers have high dimensionality to satisfy sparsity constraints, resulting in prohibitive training and inference costs.<n>Recent Mixture of Experts (MoE) approaches attempt to address this by SAEs into narrower expert networks with gated activation.<n>We propose two key innovations: (1) Multiple Expert Activation that simultaneously engages semantically weighted expert subsets to encourage specialization, and (2) Feature Scaling that enhances diversity through adaptive high-frequency scaling.
  arXiv  Detail & Related papers  (2025-11-07T22:19:34Z)
• Sycophancy Mitigation Through Reinforcement Learning with Uncertainty-Aware Adaptive Reasoning Trajectories [58.988535279557546]
  We introduce textbf sycophancy Mitigation through Adaptive Reasoning Trajectories.<n>We show that SMART significantly reduces sycophantic behavior while preserving strong performance on out-of-distribution inputs.
  arXiv  Detail & Related papers  (2025-09-20T17:09:14Z)
• Interpreting CLIP with Hierarchical Sparse Autoencoders [8.692675181549117]
  Matryoshka SAE (MSAE) learns hierarchical representations at multiple granularities simultaneously.<n>MSAE establishes a new state-of-the-art frontier between reconstruction quality and sparsity for CLIP.
  arXiv  Detail & Related papers  (2025-02-27T22:39:13Z)
• Towards Continual Learning Desiderata via HSIC-Bottleneck Orthogonalization and Equiangular Embedding [55.107555305760954]
  We propose a conceptually simple yet effective method that attributes forgetting to layer-wise parameter overwriting and the resulting decision boundary distortion. Our method achieves competitive accuracy performance, even with absolute superiority of zero exemplar buffer and 1.02x the base model.
  arXiv  Detail & Related papers  (2024-01-17T09:01:29Z)
• Accurate Neural Network Pruning Requires Rethinking Sparse Optimization [87.90654868505518]
  We show the impact of high sparsity on model training using the standard computer vision and natural language processing sparsity benchmarks. We provide new approaches for mitigating this issue for both sparse pre-training of vision models and sparse fine-tuning of language models.
  arXiv  Detail & Related papers  (2023-08-03T21:49:14Z)
• Enhancing Adversarial Training with Feature Separability [52.39305978984573]
  We introduce a new concept of adversarial training graph (ATG) with which the proposed adversarial training with feature separability (ATFS) enables to boost the intra-class feature similarity and increase inter-class feature variance. Through comprehensive experiments, we demonstrate that the proposed ATFS framework significantly improves both clean and robust performance.
  arXiv  Detail & Related papers  (2022-05-02T04:04:23Z)
• Sparsity Winning Twice: Better Robust Generalization from More Efficient Training [94.92954973680914]
  We introduce two alternatives for sparse adversarial training: (i) static sparsity and (ii) dynamic sparsity. We find both methods to yield win-win: substantially shrinking the robust generalization gap and alleviating the robust overfitting. Our approaches can be combined with existing regularizers, establishing new state-of-the-art results in adversarial training.
  arXiv  Detail & Related papers  (2022-02-20T15:52:08Z)
• Layer-wise Regularized Adversarial Training using Layers Sustainability Analysis (LSA) framework [8.701566919381223]
  An appropriate solution to adversarial attacks is adversarial training, which reaches a trade-off between robustness and generalization. This paper introduces a novel framework (Layer Sustainability Analysis) for the analysis of layer vulnerability in a given neural network in the scenario of adversarial attacks. The proposed idea performs well theoretically and experimentally for state-of-the-art multilayer perceptron and convolutional neural network architectures.
  arXiv  Detail & Related papers  (2022-02-05T20:05:53Z)
• Boosting Discriminative Visual Representation Learning with Scenario-Agnostic Mixup [54.09898347820941]
  We propose textbfScenario-textbfAgnostic textbfMixup (SAMix) for both Self-supervised Learning (SSL) and supervised learning (SL) scenarios. Specifically, we hypothesize and verify the objective function of mixup generation as optimizing local smoothness between two mixed classes. A label-free generation sub-network is designed, which effectively provides non-trivial mixup samples and improves transferable abilities.
  arXiv  Detail & Related papers  (2021-11-30T14:49:59Z)
• Robust Pre-Training by Adversarial Contrastive Learning [120.33706897927391]
  Recent work has shown that, when integrated with adversarial training, self-supervised pre-training can lead to state-of-the-art robustness. We improve robustness-aware self-supervised pre-training by learning representations consistent under both data augmentations and adversarial perturbations.
  arXiv  Detail & Related papers  (2020-10-26T04:44:43Z)

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.