Mitigating Position Bias in Transformers via Layer-Specific Positional Embedding Scaling

• URL: http://arxiv.org/abs/2606.27705v1
• Date: Fri, 26 Jun 2026 04:07:41 GMT
• Title: Mitigating Position Bias in Transformers via Layer-Specific Positional Embedding Scaling
• Abstract summary: We introduce a layer-specific positional embedding scaling(LPES) method that assigns distinct scaling factors to each layer.<n>LPES achieves a more balanced attention distribution without fine-tuning model parameters or increasing inference delay.<n>Experiments demonstrate that LPES effectively mitigates positional attention bias and delivers consistent improvements across multiple long-context benchmarks.
• Score: 47.23122393772321
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Large Language Models (LLMs) still struggle with the ``lost-in-the-middle'' problem, where critical information located in the middle of long-context inputs is often underrepresented or lost. While existing methods attempt to address this by combining multi-scale rotary position embeddings (RoPE), they typically suffer from high latency or rely on suboptimal hand-crafted scaling strategies. To overcome these limitations, we introduce a layer-specific positional embedding scaling~(LPES) method that assigns distinct scaling factors to each layer. LPES achieves a more balanced attention distribution without fine-tuning model parameters or increasing inference delay. A specially designed genetic algorithm is employed to efficiently select the optimal scaling factors for each layer by incorporating Bézier curves to significantly reduce the search space. Extensive experiments demonstrate that LPES effectively mitigates positional attention bias and delivers consistent improvements across multiple long-context benchmarks, yielding up to an $11.2$\% accuracy gain on the key-value retrieval dataset.

Related papers

• Preference Data Selection for Mitigating the Alignment Tax in Large Language Models [52.18310687523229]
  BALIGN is a balanced data selection strategy that mitigates catastrophic forgetting while optimizing alignment efficacy.<n>By aggregating three key data-centric features into a unified composite risk score, BALIGN systematically filters out high-risk preference samples.<n>Experiments on standard human preference datasets demonstrate that BALIGN strongly preserves foundational capabilities without compromising alignment gains.
  arXiv  Detail & Related papers  (2026-08-25T07:58:42Z)
• OTARo: Once Tuning for All Precisions toward Robust On-Device LLMs [21.55040910903597]
  OTARo is a novel method that enables on-device Large Language Models to flexibly switch quantization precisions.<n>It achieves consistently strong and robust performance for all precisions.
  arXiv  Detail & Related papers  (2025-11-17T08:56:27Z)
• Long-Context Modeling with Dynamic Hierarchical Sparse Attention for On-Device LLMs [17.499497967319332]
  We introduce Dynamic Hierarchical Sparse Attention (DHSA), a data-driven framework that dynamically predicts attention sparsity online without retraining.<n>DHSA matches dense attention in accuracy, while reducing prefill latency by 20-60% and peak memory usage by 35%.<n>Our experiments on Gemma2 with Needle-in-a-Haystack Test and LongBench show that DHSA matches dense attention in accuracy, while reducing prefill latency by 20-60% and peak memory usage by 35%.
  arXiv  Detail & Related papers  (2025-10-28T16:34:18Z)
• The Sparse Frontier: Sparse Attention Trade-offs in Transformer LLMs [40.35884943268004]
  We show that for very long sequences, larger and highly sparse models are preferable to smaller and dense ones.<n>There is no clear strategy that performs best across tasks and phases, with different units of sparsification or budget adaptivity needed for different scenarios.<n>We introduce and validate novel scaling laws specifically tailored for sparse attention, providing evidence that our findings are likely to hold true beyond our range of experiments.
  arXiv  Detail & Related papers  (2025-04-24T17:39:25Z)
• Layer-Specific Scaling of Positional Encodings for Superior Long-Context Modeling [26.310612987107813]
  Large language models suffer from the lost-in-the-middle'' problem, where crucial information in the middle of the context is often underrepresented or lost.<n>We propose a layer-specific positional encoding scaling method that assigns distinct scaling factors to each layer.<n>Our approach results in an average accuracy improvement of up to 20% on the Key-Value Retrieval dataset.
  arXiv  Detail & Related papers  (2025-03-06T11:59:55Z)
• Generalized EXTRA stochastic gradient Langevin dynamics [6.899153618328339]
  Langevin algorithms are popular Markov Chain Monte Carlo methods for Bayesian learning.<n>Their versions such as Langevin dynamics (SGLD) allow iterative learning based on randomly sampled mini-batches.<n>When data is decentralized across a network of agents subject to communication and privacy constraints, standard SGLD algorithms cannot be applied.
  arXiv  Detail & Related papers  (2024-12-02T21:57:30Z)
• Stability and Generalization for Distributed SGDA [70.97400503482353]
  We propose the stability-based generalization analytical framework for Distributed-SGDA. We conduct a comprehensive analysis of stability error, generalization gap, and population risk across different metrics. Our theoretical results reveal the trade-off between the generalization gap and optimization error.
  arXiv  Detail & Related papers  (2024-11-14T11:16:32Z)
• Large-scale Fully-Unsupervised Re-Identification [78.47108158030213]
  We propose two strategies to learn from large-scale unlabeled data. The first strategy performs a local neighborhood sampling to reduce the dataset size in each without violating neighborhood relationships. A second strategy leverages a novel Re-Ranking technique, which has a lower time upper bound complexity and reduces the memory complexity from O(n2) to O(kn) with k n.
  arXiv  Detail & Related papers  (2023-07-26T16:19:19Z)
• Compound Batch Normalization for Long-tailed Image Classification [77.42829178064807]
  We propose a compound batch normalization method based on a Gaussian mixture. It can model the feature space more comprehensively and reduce the dominance of head classes. The proposed method outperforms existing methods on long-tailed image classification.
  arXiv  Detail & Related papers  (2022-12-02T07:31:39Z)
• Adaptive Self-supervision Algorithms for Physics-informed Neural Networks [59.822151945132525]
  Physics-informed neural networks (PINNs) incorporate physical knowledge from the problem domain as a soft constraint on the loss function. We study the impact of the location of the collocation points on the trainability of these models. We propose a novel adaptive collocation scheme which progressively allocates more collocation points to areas where the model is making higher errors.
  arXiv  Detail & Related papers  (2022-07-08T18:17:06Z)
• Layer-adaptive sparsity for the Magnitude-based Pruning [88.37510230946478]
  We propose a novel importance score for global pruning, coined layer-adaptive magnitude-based pruning (LAMP) score. LAMP consistently outperforms popular existing schemes for layerwise sparsity selection.
  arXiv  Detail & Related papers  (2020-10-15T09:14:02Z)

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.