Improving Robustness In Sparse Autoencoders via Masked Regularization

• URL: http://arxiv.org/abs/2604.06495v1
• Date: Tue, 07 Apr 2026 21:56:23 GMT
• Title: Improving Robustness In Sparse Autoencoders via Masked Regularization
• Abstract summary: 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.
• Score: 21.265992727881514
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Sparse autoencoders (SAEs) are widely used in mechanistic interpretability to project LLM activations onto sparse latent spaces. However, sparsity alone is an imperfect proxy for interpretability, and current training objectives often result in brittle latent representations. SAEs are known to be prone to feature absorption, where general features are subsumed by more specific ones due to co-occurrence, degrading interpretability despite high reconstruction fidelity. Recent negative results on Out-of-Distribution (OOD) performance further underscore broader robustness related failures tied to under-specified training objectives. We address this by proposing a masking-based regularization that randomly replaces tokens during training to disrupt co-occurrence patterns. This improves robustness across SAE architectures and sparsity levels reducing absorption, enhancing probing performance, and narrowing the OOD gap. Our results point toward a practical path for more reliable interpretability tools.

Related papers

• When Pruning Meets Interpretability: Preserving Sparse Autoencoder Robustness in LLMs [15.418870425179875]
  Sparse autoencoders (SAEs) are widely used to interpret the internal representations of large language models.<n>We show that for a fixed SAE, its impact is governed by perturbation energy.<n>We propose a layer-wise sparsity allocation strategy, achieving lower perplexity under the same average pruning sparsity.
  arXiv  Detail & Related papers  (2026-08-26T15:57:42Z)
• From Sparse Features to Trustworthy Proxies: Certifying SAE-Based Interpretability [53.681678236115836]
  We study when can an SAE-based explanation be treated as a faithful view of an underlying frozen LM.<n>Our framework derives an upper bound on the base model's expected risk using four measurable quantities.<n>We show that the bound becomes non-vacuous on GPT-2 Small, Gemma-2B, and Llama-3-8B at practical sample sizes.
  arXiv  Detail & Related papers  (2026-06-16T18:28:23Z)
• What Does the Server See? Understanding Privacy Leakage from Large Language Models in Split Inference [13.194874392306085]
  We introduce ActInv, which solves an intermediate activation matching problem to reconstruct the client's input.<n>To understand this vulnerability, we develop Perturbation Amplification Factor (PAF), a metric for quantifying a layer's inherent resistance to reconstruction.<n>Our analysis reveals that privacy vulnerability is not uniform across layers, with some layers being highly susceptible to leakage while others offer natural resistance.
  arXiv  Detail & Related papers  (2026-05-22T02:14:16Z)
• Self-Distilled Trajectory-Aware Boltzmann Modeling: Bridging the Training-Inference Discrepancy in Diffusion Language Models [65.89572755202245]
  Diffusion Language Models (DLMs) offer stronger global awareness and highly parallel generation.<n>Post-training DLMs with standard Negative Evidence Lower Bound (NELBO)-based supervised fine-tuning remains inefficient.<n>We propose a self-distilled trajectory-based post-training framework that aligns training with the easy-to-hard structure of inference.
  arXiv  Detail & Related papers  (2026-05-12T09:39:06Z)
• When Does Sparsity Mitigate the Curse of Depth in LLMs [53.137717161619484]
  We show that sparsity acts as a regulator of variance propagation and thereby improves depth utilization.<n>Our results reveal sparsity as a key yet previously overlooked mechanism for effective depth scaling in large language models.
  arXiv  Detail & Related papers  (2026-03-16T15:04:16Z)
• SSA: Sparse Sparse Attention by Aligning Full and Sparse Attention Outputs in Feature Space [31.792742191540356]
  We propose SSA (Sparse Sparse Attention), a unified training framework that considers both sparse and full attention.<n> SSA achieves state-of-the-art performance under both sparse and full attention inference.<n>We show that native sparse-attention training surprisingly improves long-context by mitigating the over-allocation of attention values in sink areas.
  arXiv  Detail & Related papers  (2025-11-25T09:21:57Z)
• Shortcut Invariance: Targeted Jacobian Regularization in Disentangled Latent Space [7.8904984750896885]
  Deep neural networks are prone to learning shortcuts, spurious and easily learned correlations.<n>We present a simple and effective training method that renders the classifier functionally invariant to shortcut signals.<n>We analyze this as targeted Jacobian regularization, which forces the classifier to ignore spurious features and rely on more complex, core semantic signals.
  arXiv  Detail & Related papers  (2025-11-24T07:09:08Z)
• 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)
• Train One Sparse Autoencoder Across Multiple Sparsity Budgets to Preserve Interpretability and Accuracy [3.2248482136498433]
  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.
  arXiv  Detail & Related papers  (2025-05-30T11:20:44Z)
• Improving LLM Safety Alignment with Dual-Objective Optimization [81.98466438000086]
  Existing training-time safety alignment techniques for large language models (LLMs) remain vulnerable to jailbreak attacks.<n>We propose an improved safety alignment that disentangles DPO objectives into two components: (1) robust refusal training, which encourages refusal even when partial unsafe generations are produced, and (2) targeted unlearning of harmful knowledge.
  arXiv  Detail & Related papers  (2025-03-05T18:01:05Z)
• Augmenting Unsupervised Reinforcement Learning with Self-Reference [63.68018737038331]
  Humans possess the ability to draw on past experiences explicitly when learning new tasks. We propose the Self-Reference (SR) approach, an add-on module explicitly designed to leverage historical information. Our approach achieves state-of-the-art results in terms of Interquartile Mean (IQM) performance and Optimality Gap reduction on the Unsupervised Reinforcement Learning Benchmark.
  arXiv  Detail & Related papers  (2023-11-16T09:07:34Z)
• Rethinking Uncertainty in Deep Learning: Whether and How it Improves Robustness [20.912492996647888]
  adversarial training (AT) suffers from poor performance both on clean examples and under other types of attacks. Regularizers that encourage uncertain outputs, such as entropy (EntM) and label smoothing (LS) can maintain accuracy on clean examples and improve performance under weak attacks. In this paper, we revisit uncertainty promotion regularizers, including EntM and LS, in the field of adversarial learning.
  arXiv  Detail & Related papers  (2020-11-27T03:22:50Z)

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.