SSA: Sparse Sparse Attention by Aligning Full and Sparse Attention Outputs in Feature Space

• URL: http://arxiv.org/abs/2511.20102v1
• Date: Tue, 25 Nov 2025 09:21:57 GMT
• Title: SSA: Sparse Sparse Attention by Aligning Full and Sparse Attention Outputs in Feature Space
• Abstract summary: 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.
• Score: 31.792742191540356
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: The quadratic complexity of full attention limits efficient long-context processing in large language models (LLMs). Sparse attention mitigates this cost by restricting each query to attend to a subset of previous tokens; however, training-free approaches often lead to severe performance degradation. Native sparse-attention methods (e.g., NSA, MoBA) alleviate this issue, yet exhibit a critical paradox: they produce lower attention sparsity than full-attention models, despite aiming to approximate full attention, which may constrain their effectiveness. We attribute this paradox to gradient update deficiency: low-ranked key-value pairs excluded during sparse training receive neither forward contribution nor backward gradients, and thus never learn proper suppression. To overcome this limitation, we propose SSA (Sparse Sparse Attention), a unified training framework that considers both sparse and full attention and enforces bidirectional alignment at every layer. This design preserves gradient flow to all tokens while explicitly encouraging sparse-attention outputs to align with their full-attention counterparts, thereby promoting stronger sparsity. As a result, SSA achieves state-of-the-art performance under both sparse and full attention inference across multiple commonsense benchmarks. Furthermore, SSA enables models to adapt smoothly to varying sparsity budgets; performance improves consistently as more tokens are allowed to attend, supporting flexible compute-performance trade-offs at inference time. Finally, we show that native sparse-attention training surprisingly improves long-context extrapolation by mitigating the over-allocation of attention values in sink areas, with SSA demonstrating the strongest extrapolation capability.

Related papers

• Disentangling Semantic Attention from Structural Bias in the Attention Manifold [61.68044165974505]
  Multimodal Large Language Models (MLLMs) exhibit disproportionate attention toward certain semantically uninformative visual tokens.<n>We introduce Saliency-guided Purification and Adaptive Redistribution (SPAR), a training-free, plug-and-play intervention.
  arXiv  Detail & Related papers  (2026-07-27T05:30:19Z)
• See Only When Needed: Context-Aware Attention Intervention for Mitigating Hallucinations in LVLMs [81.71384150441163]
  We propose Context-aware Attention Intervention (CAI) for large vision-language models (LVLMs)<n>CAI enforces a see only when needed principle via two-axis selectivity: where to look and when to intervene.<n> experiments across multiple LVLM backbones and benchmarks show that CAI achieves state-of-the-art hallucination mitigation.
  arXiv  Detail & Related papers  (2026-06-29T06:35:47Z)
• Self-Consistent Latent Reasoning: Long Latent Sequence Reasoning for Vision-Language Model [56.21523258053447]
  SCOLAR (Self-COnsistent LAtent Reasoning) introduces a lightweight detransformer that generates auxiliary visual tokens in a single shot.<n>SCOLAR extends acceptable latent CoT length by over $30times$, achieves state-of-the-art among open-source models on real-world reasoning benchmarks.
  arXiv  Detail & Related papers  (2026-05-12T14:13:08Z)
• 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)
• Beyond Attention Magnitude: Leveraging Inter-layer Rank Consistency for Efficient Vision-Language-Action Models [75.27089289058613]
  We introduce textbfTIES (textbfTau-guided textbfInter-layer textbfEfficient textbfSelection), a dynamic framework guided by inter-layer token ranking consistency.<n>On the CogACT + SIMPLER benchmark, TIES improves average success rates by 6% while reducing token usage by 78%, and demonstrate strong generalization across diverse decoders and benchmarks.
  arXiv  Detail & Related papers  (2026-03-26T02:13:03Z)
• ASAP: Attention-Shift-Aware Pruning for Efficient LVLM Inference [26.71949723961501]
  We propose ASAP, a novel training-free, KV-Cache-compatible pruning recipe.<n>We mitigate the attention shift by utilizing a dynamic bidirectional soft attention mask.<n>We also posit that high semantic redundancy within the token set degrades performance.
  arXiv  Detail & Related papers  (2026-03-15T18:51:31Z)
• Look Carefully: Adaptive Visual Reinforcements in Multimodal Large Language Models for Hallucination Mitigation [51.743225614196774]
  Multimodal large language models (MLLMs) have achieved remarkable progress in vision-language reasoning.<n>They remain vulnerable to hallucination, where generated content deviates from visual evidence.<n>Recent vision enhancement methods attempt to address this issue by reinforcing visual tokens during decoding.<n>We propose Adaptive Visual Reinforcement (AIR), a training-free framework for MLLMs.
  arXiv  Detail & Related papers  (2026-02-27T14:18:51Z)
• Rethinking Multi-Condition DiTs: Eliminating Redundant Attention via Position-Alignment and Keyword-Scoping [61.459927600301654]
  Multi-condition control is bottlenecked by the conventional concatenate-and-attend'' strategy.<n>Our analysis reveals that much of this cross-modal interaction is spatially or semantically redundant.<n>We propose Position-aligned and Keyword-scoped Attention (PKA), a highly efficient framework designed to eliminate these redundancies.
  arXiv  Detail & Related papers  (2026-02-06T16:39:10Z)
• Accordion-Thinking: Self-Regulated Step Summaries for Efficient and Readable LLM Reasoning [62.680551162054975]
  We introduce an end-to-end framework where LLMs learn to self-regulate the granularity of the reasoning steps through dynamic summarization.<n>We apply reinforcement learning to incentivize this capability further, uncovering a critical insight: the accuracy gap between the highly efficient Fold mode and the exhaustive Unfold mode progressively narrows.<n>Our Accordion-Thinker demonstrates that with learned self-compression, LLMs can tackle complex reasoning tasks with minimal dependency token overhead.
  arXiv  Detail & Related papers  (2026-02-03T08:34:20Z)
• Punctuation-aware Hybrid Trainable Sparse Attention for Large Language Models [44.28116882776357]
  We present textbfPunctuation-aware textbfHybrid textbfSparse textbfAttention textbf(PHSA), a trainable sparse attention framework that leverages punctuation tokens as semantic boundary anchors.<n>Specifically, we design a dual-branch aggregation mechanism that fuses global semantic representations with punctuation-enhanced boundary features, preserving the core semantic structure while introducing almost no additional computational overhead.
  arXiv  Detail & Related papers  (2026-01-06T08:47:16Z)
• Training-free Context-adaptive Attention for Efficient Long Context Modeling [57.703159205740185]
  Training-free Context-adaptive Attention (TCA-Attention) is a training-free sparse attention mechanism that selectively attends to only the informative tokens for efficient long-context inference.<n>TCA-Attention achieves a 2.8$times$ speedup and reduces KV cache by 61% at 128K context length while maintaining performance comparable to full attention.
  arXiv  Detail & Related papers  (2025-12-10T01:54:57Z)
• FrugalPrompt: Reducing Contextual Overhead in Large Language Models via Token Attribution [3.4666771782038652]
  Large language models (LLMs) owe much of their stellar performance to expansive input contexts, yet such verbosity inflates monetary costs, carbon footprint, and inference-time latency.<n>We introduce FrugalPrompt, a novel prompt compression framework for LLMs, which retains only the most semantically significant tokens.<n>We evaluate the approach across four NLP tasks: Sentiment Analysis, Commonsense QA, Summarization, and Mathematical Reasoning.
  arXiv  Detail & Related papers  (2025-10-18T10:22:13Z)
• Don't Just Chase "Highlighted Tokens" in MLLMs: Revisiting Visual Holistic Context Retention [50.97683288777336]
  Multimodal Large Language Models (MLLMs) suffer from considerable computational overhead due to their reliance on massive visual tokens.<n>Recent studies have explored token pruning to alleviate this problem, which typically uses text-vision cross-attention.<n>We propose HoloV, a plug-and-play visual token pruning framework for efficient inference.
  arXiv  Detail & Related papers  (2025-10-03T11:33:40Z)
• Think Clearly: Improving Reasoning via Redundant Token Pruning [57.01254508252785]
  We show that deliberately removing redundancy in the reasoning process significantly improves performance.<n>We demonstrate that our method significantly improves overall accuracy across reasoning-intensive benchmarks without any training.
  arXiv  Detail & Related papers  (2025-06-17T06:04:01Z)
• STAR: Stage-Wise Attention-Guided Token Reduction for Efficient Large Vision-Language Models Inference [3.9464481148889354]
  We propose STAR (Stage-wise Attention-guided token Reduction), a training-free, plug-and-play framework that approaches token pruning from a global perspective.<n>Instead of pruning at a single point, STAR performs attention-guided reduction in two complementary stages: an early-stage pruning based on visual self-attention to remove redundant low-level features, and a later-stage pruning guided by cross-modal attention to discard task-irrelevant tokens.
  arXiv  Detail & Related papers  (2025-05-18T10:44:45Z)
• PowerAttention: Exponentially Scaling of Receptive Fields for Effective Sparse Attention [73.26995918610669]
  Large Language Models (LLMs) face efficiency bottlenecks due to the quadratic complexity of the attention mechanism when processing long contexts.<n>We introduce PowerAttention, a novel sparse attention design that facilitates effective and complete context extension.<n>Experiments demonstrate that PowerAttention outperforms existing static sparse attention methods by $5sim 40%$.
  arXiv  Detail & Related papers  (2025-03-05T15:24:11Z)
• Tactic: Adaptive Sparse Attention with Clustering and Distribution Fitting for Long-Context LLMs [10.52833484759311]
  We propose Tactic, a sparsity-adaptive and calibration-free sparse attention mechanism.<n>It dynamically selects tokens based on their cumulative attention scores rather than a fixed token budget.<n>We show that Tactic outperforms existing sparse attention algorithms, achieving superior accuracy and up to 7.29x decode attention speedup.
  arXiv  Detail & Related papers  (2025-02-17T08:39:43Z)
• SeerAttention: Learning Intrinsic Sparse Attention in Your LLMs [10.702409298302547]
  SeerAttention learns the block-level attention sparsity from the Large Language Models itself.<n>Inspired by the gating mechanism in Mixture of Experts (MoE), SeerAttention augments the conventional attention with a learnable gate.<n>Our evaluation results demonstrate that SeerAttention achieves better model accuracy and lower latency for long-context pre-filling.
  arXiv  Detail & Related papers  (2024-10-17T07:07:09Z)
• Unveiling and Harnessing Hidden Attention Sinks: Enhancing Large Language Models without Training through Attention Calibration [15.36841874118801]
  We aim to provide a more profound understanding of the existence of attention sinks within large language models (LLMs) We propose a training-free Attention Technique (ACT) that automatically optimize the attention distributions on the fly during inference in an input-adaptive manner. ACT achieves an average improvement of up to 7.30% in accuracy across different datasets when applied to Llama-30B.
  arXiv  Detail & Related papers  (2024-06-22T07:00:43Z)
• Calibrating Undisciplined Over-Smoothing in Transformer for Weakly Supervised Semantic Segmentation [51.14107156747967]
  Weakly supervised semantic segmentation (WSSS) has attracted considerable attention because it requires fewer annotations than fully supervised approaches.<n>We propose an Adaptive Re-Activation Mechanism (AReAM) to control deep-level attention to undisciplined over-smoothing.<n>AReAM substantially improves segmentation performance compared with existing WSSS methods, reducing noise while sharpening focus on relevant semantic regions.
  arXiv  Detail & Related papers  (2023-05-04T19:11:33Z)
• Flip Learning: Erase to Segment [65.84901344260277]
  Weakly-supervised segmentation (WSS) can help reduce time-consuming and cumbersome manual annotation. We propose a novel and general WSS framework called Flip Learning, which only needs the box annotation. Our proposed approach achieves competitive performance and shows great potential to narrow the gap between fully-supervised and weakly-supervised learning.
  arXiv  Detail & Related papers  (2021-08-02T09:56:10Z)

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.