Research map / Architectures and efficiency / Attention and transformers
Probing Rotary Position Embeddings through Frequency Entropy
Yui Oka, Kentaro Hanafusa, Taku Hasegawa, Kyosuke Nishida, Kuniko Saito
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 changed | Where | Measured | Effect |
|---|---|---|---|
| attenuate high-SpectrumFE outlier and low-SequenceFE periodic rotary pairs (alpha=0.1) (ablation) | Llama-4 17B · MMLU | accuracy: 60.05 -> 60.81 | No clear effect |
| attenuate outlier and periodic rotary pairs (ablation) | Llama-4 17B · TruthfulQA | accuracy: 97.32 -> 97.99 | No clear effect |
| attenuate outlier and periodic rotary pairs (ablation) | Llama-3 8B · HellaSwag, TruthfulQA, MMLU | accuracy: 60.16 -> 60.16 | No clear effect |
| attenuate outlier and periodic rotary pairs (ablation) | Qwen3 8B · HellaSwag | accuracy: 58.94 -> 58.92 | No clear effect |
| attenuate outlier and periodic rotary pairs (ablation) | Gemma-2 9B · HellaSwag | accuracy: 61.02 -> 61.21 | No clear effect |
| attenuate low-SpectrumFE (tau<0.2) band dimensions (ablation) | LLMs with RoPE · WikiText-103 | perplexity | Hurt |
| attenuate high-SpectrumFE (tau>0.4) outlier dimensions (ablation) | LLMs with RoPE · WikiText-103 | perplexity | No clear effect |
| attenuate low-SequenceFE (tau<0.6) periodic dimensions (ablation) | LLMs with RoPE · WikiText-103 | perplexity | Improved |
Closest papers by meaning
- Round and Round We Go! What makes Rotary Positional Encodings useful? (ICLR 2025 · similarity 0.88)
- Dynamical Properties of Tokens in Self-Attention and Effects of Positional Encoding (NeurIPS 2025 · similarity 0.88)
- What Rotary Position Embedding Can Tell Us: Identifying Query and Key Weights Corresponding to Basic Syntactic or High-level Semantic Information (NeurIPS 2024 · similarity 0.87)
- ComRoPE: Scalable and Robust Rotary Position Embedding Parameterized by Trainable Commuting Angle Matrices (CVPR 2025 · similarity 0.86)
- A Circular Argument: Does RoPE need to be Equivariant for Vision? (NeurIPS 2025 · similarity 0.86)
- LieRE: Lie Rotational Positional Encodings (ICML 2025 · similarity 0.86)
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.
