Crisp Attention: Regularizing Transformers via Structured Sparsity

• URL: http://arxiv.org/abs/2508.06016v1
• Date: Fri, 08 Aug 2025 05:04:28 GMT
• Title: Crisp Attention: Regularizing Transformers via Structured Sparsity
• Abstract summary: We introduce structured, post-hoc sparsity to the attention mechanism of a DistilBERT model during fine-tuning on the SST-2 sentiment analysis task.<n>Our model with 80% attention sparsity achieves a validation accuracy of 91.59%, a 0.97% absolute improvement over the dense baseline.
• Score: 0.29998889086656577
• License: http://creativecommons.org/licenses/by-sa/4.0/
• Abstract: The quadratic computational cost of the self-attention mechanism is a primary challenge in scaling Transformer models. While attention sparsity is widely studied as a technique to improve computational efficiency, it is almost universally assumed to come at the cost of model accuracy. In this paper, we report a surprising counter-example to this common wisdom. By introducing structured, post-hoc sparsity to the attention mechanism of a DistilBERT model during fine-tuning on the SST-2 sentiment analysis task, we find that model accuracy improves significantly. Our model with 80\% attention sparsity achieves a validation accuracy of 91.59\%, a 0.97\% absolute improvement over the dense baseline. We hypothesize that this phenomenon is due to sparsity acting as a powerful implicit regularizer, preventing the model from overfitting by forcing it to make predictions with a more constrained and robust set of features. Our work recasts attention sparsity not just as a tool for computational efficiency, but as a potential method for improving the generalization and performance of Transformer models.

Related papers

• From Fake Focus to Real Precision: Confusion-Driven Adversarial Attention Learning in Transformers [0.0]
  Transformer-based models have been widely adopted for sentiment analysis tasks due to their exceptional ability to capture contextual information.<n>We observe that existing models tend to allocate attention primarily to common words, overlooking less popular yet highly task-relevant terms.<n>We propose an Adversarial Feedback for Attention(AFA) training mechanism that enables the model to automatically redistribute attention weights to appropriate focal points.
  arXiv  Detail & Related papers  (2025-12-19T01:48:25Z)
• STORE: Semantic Tokenization, Orthogonal Rotation and Efficient Attention for Scaling Up Ranking Models [11.965535230928372]
  Store is a unified and scalable token-based ranking framework built upon three core innovations.<n>Our framework consistently improves prediction accuracy(online CTR by 2.71%, AUC by 1.195%) and training effeciency (1.84 throughput)
  arXiv  Detail & Related papers  (2025-11-24T06:20:02Z)
• Transformers Learn Faster with Semantic Focus [57.97235825738412]
  We study sparse transformers in terms of learnability and generalization.<n>We find that input-dependent sparse attention models appear to converge faster and generalize better than standard attention models.
  arXiv  Detail & Related papers  (2025-06-17T01:19:28Z)
• Does Scaling Law Apply in Time Series Forecasting? [2.127584662240465]
  We propose Alinear, an ultra-lightweight forecasting model that achieves competitive performance using only k-level parameters.<n>Experiments on seven benchmark datasets demonstrate that Alinear consistently outperforms large-scale models.<n>This work challenges the prevailing belief that larger models are inherently better and suggests a paradigm shift toward more efficient time series modeling.
  arXiv  Detail & Related papers  (2025-05-15T11:04:39Z)
• State Tuning: State-based Test-Time Scaling on RWKV-7 [0.747193191854175]
  We introduce a novel state-based approach to test-time scaling, which we term state tuning, tailored to the RNN-based RWKV-7 model.<n>By exploiting the unique strengths of RWKV-7, our method achieves state-of-the-art performance on the target task without altering the model's pre-trained weights.
  arXiv  Detail & Related papers  (2025-04-07T14:04:30Z)
• Transformer Meets Twicing: Harnessing Unattended Residual Information [2.1605931466490795]
  Transformer-based deep learning models have achieved state-of-the-art performance across numerous language and vision tasks.<n>While the self-attention mechanism has proven capable of handling complex data patterns, it has been observed that the representational capacity of the attention matrix degrades significantly across transformer layers.<n>We propose the Twicing Attention, a novel attention mechanism that uses kernel twicing procedure in nonparametric regression to alleviate the low-pass behavior of associated NLM smoothing.
  arXiv  Detail & Related papers  (2025-03-02T01:56:35Z)
• Top-Theta Attention: Sparsifying Transformers by Compensated Thresholding [1.6112718683989882]
  We introduce Top-Theta Attention, or simply Top-$theta$, which selectively prunes less essential attention elements by comparing them against carefully calibrated thresholds.<n>This method greatly improves the efficiency of self-attention matrix multiplication while preserving model accuracy.<n>Unlike top-k attention, Top-$theta$ eliminates full-vector dependency, making it suitable for tiling and scale-out and avoiding costly top-k search.
  arXiv  Detail & Related papers  (2025-02-12T12:50:15Z)
• Enhancing Transformer-based models for Long Sequence Time Series Forecasting via Structured Matrix [7.3758245014991255]
  Self-attention mechanism as the core component of Transformer-based models exhibits great potential.<n>We propose a novel architectural framework that enhances Transformer-based models through the integration of Surrogate Attention Blocks (SAB) and Surrogate Feed-Forward Neural Network Blocks (SFB)<n>The framework reduces both time and space complexity by the replacement of the self-attention and feed-forward layers with SAB and SFB.
  arXiv  Detail & Related papers  (2024-05-21T02:37:47Z)
• TransNormerLLM: A Faster and Better Large Language Model with Improved TransNormer [34.790081960470964]
  We present TransNormerLLM, the first linear attention-based Large Language Model (LLM) We make advanced modifications that include positional embedding, linear attention acceleration, gating mechanisms, tensor normalization, and inference acceleration and stabilization. We validate our model design through a series of ablations and train models with sizes of 385M, 1B, and 7B on our self-collected corpus.
  arXiv  Detail & Related papers  (2023-07-27T16:45:33Z)
• Precision-Recall Divergence Optimization for Generative Modeling with GANs and Normalizing Flows [54.050498411883495]
  We develop a novel training method for generative models, such as Generative Adversarial Networks and Normalizing Flows. We show that achieving a specified precision-recall trade-off corresponds to minimizing a unique $f$-divergence from a family we call the textitPR-divergences. Our approach improves the performance of existing state-of-the-art models like BigGAN in terms of either precision or recall when tested on datasets such as ImageNet.
  arXiv  Detail & Related papers  (2023-05-30T10:07:17Z)
• How Much Does Attention Actually Attend? Questioning the Importance of Attention in Pretrained Transformers [59.57128476584361]
  We introduce PAPA, a new probing method that replaces the input-dependent attention matrices with constant ones. We find that without any input-dependent attention, all models achieve competitive performance. We show that better-performing models lose more from applying our method than weaker models, suggesting that the utilization of the input-dependent attention mechanism might be a factor in their success.
  arXiv  Detail & Related papers  (2022-11-07T12:37:54Z)
• Powerpropagation: A sparsity inducing weight reparameterisation [65.85142037667065]
  We introduce Powerpropagation, a new weight- parameterisation for neural networks that leads to inherently sparse models. Models trained in this manner exhibit similar performance, but have a distribution with markedly higher density at zero, allowing more parameters to be pruned safely. Here, we combine Powerpropagation with a traditional weight-pruning technique as well as recent state-of-the-art sparse-to-sparse algorithms, showing superior performance on the ImageNet benchmark.
  arXiv  Detail & Related papers  (2021-10-01T10:03:57Z)
• Efficient pre-training objectives for Transformers [84.64393460397471]
  We study several efficient pre-training objectives for Transformers-based models. We prove that eliminating the MASK token and considering the whole output during the loss are essential choices to improve performance.
  arXiv  Detail & Related papers  (2021-04-20T00:09:37Z)
• Attention that does not Explain Away [54.42960937271612]
  Models based on the Transformer architecture have achieved better accuracy than the ones based on competing architectures for a large set of tasks. A unique feature of the Transformer is its universal application of a self-attention mechanism, which allows for free information flow at arbitrary distances. We propose a doubly-normalized attention scheme that is simple to implement and provides theoretical guarantees for avoiding the "explaining away" effect.
  arXiv  Detail & Related papers  (2020-09-29T21:05:39Z)

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.