Untwisting RoPE: Frequency Control for Shared Attention in DiTs

• URL: http://arxiv.org/abs/2602.05013v1
• Date: Wed, 04 Feb 2026 20:01:59 GMT
• Title: Untwisting RoPE: Frequency Control for Shared Attention in DiTs
• Authors: Aryan Mikaeili, Or Patashnik, Andrea Tagliasacchi, Daniel Cohen-Or, Ali Mahdavi-Amiri,
• Abstract summary: Positional encodings are essential to transformer-based generative models.<n>We show that Rotary Positional Embeddings (RoPE) naturally decomposes into frequency components with distinct positional sensitivities.<n>We introduce a method for selectively modulating RoPE frequency bands so that attention reflects semantic similarity rather than strict positional alignment.
• Score: 84.14005261938284
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Positional encodings are essential to transformer-based generative models, yet their behavior in multimodal and attention-sharing settings is not fully understood. In this work, we present a principled analysis of Rotary Positional Embeddings (RoPE), showing that RoPE naturally decomposes into frequency components with distinct positional sensitivities. We demonstrate that this frequency structure explains why shared-attention mechanisms, where a target image is generated while attending to tokens from a reference image, can lead to reference copying, in which the model reproduces content from the reference instead of extracting only its stylistic cues. Our analysis reveals that the high-frequency components of RoPE dominate the attention computation, forcing queries to attend mainly to spatially aligned reference tokens and thereby inducing this unintended copying behavior. Building on these insights, we introduce a method for selectively modulating RoPE frequency bands so that attention reflects semantic similarity rather than strict positional alignment. Applied to modern transformer-based diffusion architectures, where all tokens share attention, this modulation restores stable and meaningful shared attention. As a result, it enables effective control over the degree of style transfer versus content copying, yielding a proper style-aligned generation process in which stylistic attributes are transferred without duplicating reference content.

Related papers

• ROAP: A Reading-Order and Attention-Prior Pipeline for Optimizing Layout Transformers in Key Information Extraction [5.594845708011402]
  This paper presents ROAP, a lightweight and architecture-agnostic pipeline designed to optimize attention distributions in layout Transformers.<n> experiments on the FUNSD and CORD benchmarks demonstrate that ROAP consistently improves the performance of backbones.
  arXiv  Detail & Related papers  (2026-01-09T02:02:37Z)
• Do traveling waves make good positional encodings? [44.55744608160896]
  We propose RollPE, a novel positional encoding mechanism based on traveling waves.<n>We show it significantly outperforms traditional absolute positional embeddings.<n>We derive a mathematical equivalence of RollPE to a particular configuration of RoPE.
  arXiv  Detail & Related papers  (2025-11-11T14:32:45Z)
• Context-aware Rotary Position Embedding [0.0]
  Rotary Positional Embeddings (RoPE) have become a widely adopted solution due to their compatibility with relative position encoding and computational efficiency.<n>We propose CARoPE (Context-Aware Rotary Positional Embedding), a novel generalization of RoPE that dynamically generates head-specific frequency patterns conditioned on token embeddings.<n>CaroPE consistently outperforms RoPE and other common positional encoding baselines, achieving significantly lower perplexity, even at longer context lengths.
  arXiv  Detail & Related papers  (2025-07-30T20:32:19Z)
• Revisiting LRP: Positional Attribution as the Missing Ingredient for Transformer Explainability [53.21677928601684]
  Layer-wise relevance propagation is one of the most promising approaches to explainability in deep learning.<n>We propose specialized theoretically-grounded LRP rules designed to propagate attributions across various positional encoding methods.<n>Our method significantly outperforms the state-of-the-art in both vision and NLP explainability tasks.
  arXiv  Detail & Related papers  (2025-06-02T18:07:55Z)
• PaTH Attention: Position Encoding via Accumulating Householder Transformations [56.32365080761523]
  PaTH is a flexible data-dependent position encoding scheme based on accumulated products of Householder transformations.<n>We derive an efficient parallel algorithm for training through exploiting a compact representation of products of Householder matrices.
  arXiv  Detail & Related papers  (2025-05-22T08:36:09Z)
• Unpacking Positional Encoding in Transformers: A Spectral Analysis of Content-Position Coupling [10.931433906211534]
  Positional encoding (PE) is essential for enabling Transformers to model sequential structure.<n>We present a unified framework that analyzes PE through the spectral properties of Toeplitz and related matrices.<n>We establish explicit content-relative mixing with relative-position Toeplitz signals as a key principle for effective PE design.
  arXiv  Detail & Related papers  (2025-05-19T12:11:13Z)
• Beyond Position: the emergence of wavelet-like properties in Transformers [6.552700667389349]
  We show that attention heads evolve to implement multi-resolution processing analogous to wavelet transforms.<n>Our findings suggest that the effectiveness of modern Transformers stems from their remarkable ability to spontaneously develop optimal, multi-resolution decompositions.
  arXiv  Detail & Related papers  (2024-10-23T17:48:28Z)
• Efficient Diffusion Transformer with Step-wise Dynamic Attention Mediators [83.48423407316713]
  We present a novel diffusion transformer framework incorporating an additional set of mediator tokens to engage with queries and keys separately. Our model initiates the denoising process with a precise, non-ambiguous stage and gradually transitions to a phase enriched with detail. Our method achieves a state-of-the-art FID score of 2.01 when integrated with the recent work SiT.
  arXiv  Detail & Related papers  (2024-08-11T07:01:39Z)
• Locality-Aware Generalizable Implicit Neural Representation [54.93702310461174]
  Generalizable implicit neural representation (INR) enables a single continuous function to represent multiple data instances. We propose a novel framework for generalizable INR that combines a transformer encoder with a locality-aware INR decoder. Our framework significantly outperforms previous generalizable INRs and validates the usefulness of the locality-aware latents for downstream tasks.
  arXiv  Detail & Related papers  (2023-10-09T11:26:58Z)
• Conformer-based End-to-end Speech Recognition With Rotary Position Embedding [11.428057887454008]
  We introduce rotary position embedding (RoPE) in the convolution-augmented transformer (conformer) RoPE encodes absolute positional information into the input sequence by a rotation matrix, and then naturally incorporates explicit relative position information into a self-attention module. Our model achieves a relative word error rate reduction of 8.70% and 7.27% over the conformer on test-clean and test-other sets of the LibriSpeech corpus respectively.
  arXiv  Detail & Related papers  (2021-07-13T08:07:22Z)
• Rethinking Global Context in Crowd Counting [70.54184500538338]
  A pure transformer is used to extract features with global information from overlapping image patches. Inspired by classification, we add a context token to the input sequence, to facilitate information exchange with tokens corresponding to image patches.
  arXiv  Detail & Related papers  (2021-05-23T12:44:27Z)

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.