AI Paper Atlas

Research map / Probabilistic and scientific modeling / Molecules and proteins

nD-RoPE: A Generalized RoPE for n-Dimensional Position Embedding

Boyang Li, Yulin Wu, Sizhe Xu, Nuoxian Huang, Zhonghang Yuan, Shangyi Guo, Shu Yang, Takahiro Yabe

ICML 2026 · Probabilistic and scientific modeling › Molecules and proteins · approach: Equivariant and disentangled latent representations · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use nD-RoPE instead of RoPE-Mixed + APEViT-S · ImageNet-1K 224top-1 accuracy (%): 80.92 -> 81.07Improved
use nD-RoPE instead of RoPE-Mixed + APE under 30 degree rotationViT-S · ImageNet-1K rotated 30 degreestop-1 accuracy (%): 71.12 -> 78.51Improved
use nD-RoPE instead of RoPE-Mixed + APE at 0 degree rotationViT-S · ImageNet-1K unrotatedtop-1 accuracy (%): 81.13 -> 80.81No clear effect
add YaRN to nD-RoPE at 1024 resolutionnD-RoPE ViT-S · ImageNet-1K eval 1024top-1 accuracy (%): 35.51 -> 68.46Improved
add YaRN to RoPE-Mixed at 1024 resolutionRoPE-Mixed ViT-S · ImageNet-1K eval 1024top-1 accuracy (%): 16.63 -> 43.48Improved
use nD-RoPE with vector attention instead of dot-product attentionPoint Transformer · ModelNet40 2048 pointsinstance mIoU (%): 85.07 -> 85.97Improved
use nD-RoPE with vector attention instead of dot-product attentionPoint Transformer v2 · SemanticKITTI grid 0.05instance mIoU (%): 67.12 -> 71.91Improved
use nD-RoPE instead of learnable relative PEPoint Transformer · ModelNet40 2048 pointsinstance mIoU (%): 82.58 -> 85.97Improved
use balanced scales per head (6 scales x 10 heads) instead of 64 scales x 1 head (ablation)nD-RoPE · ImageNet-1K 224top-1 accuracy (%): 70.99 -> 81.07Improved
use balanced scales per head (6 scales x 10 heads) instead of 1 scale x 64 heads (ablation)nD-RoPE · ImageNet-1K 224top-1 accuracy (%): 65.93 -> 81.07Improved
use frequency base 100 instead of 10^4 (ablation)nD-RoPE vector attention · ModelNet40 256 pointsinstance mIoU (%): 45.94 -> 55.04Improved
use frequency base 100 instead of 10^6 (ablation)nD-RoPE vector attention · ModelNet40 1536 pointsinstance mIoU (%): 82.67 -> 85.61Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy