AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Spectral Attention Steering for Prompt Highlighting

Weixian Waylon Li, Yuchen Niu, Yongxin Yang, Keshuang Li, Tiejun Ma, Shay B Cohen

ICLR 2026 · Architectures and efficiency › Attention and transformers · approach: Sparse and structured attention · 10 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
edit key embeddings with SEKA instead of post-hoc attention editing (PASTA)Qwen3-4B · CounterFact ESefficacy score: 97.16 -> 99.02Improved
use SEKA instead of PASTAQwen3-14B · Pronoun A. P. Scoreall-changed P. score: 90.94 -> 98.54Improved
use random projections (w/o learn) instead of spectrally learned ones (ablation)SEKA · CounterFact ES (Qwen3-4B)efficacy score: 99.02 -> 94.96Hurt
remove learned projections and KV-head filtering (ablation)SEKA · Pronoun A. P. Score (Qwen3-4B)all-changed P. score: 93.26 -> 36.95Hurt
add query-adaptive expert routing (AdaSEKA) to SEKASEKA · Pronoun A. P. Score (Gemma3-4B)all-changed P. score: 81.11 -> 93.58Improved
AdaSEKA instead of SEKASEKA · CounterFact ES (Gemma3-12B)efficacy score: 98.86 -> 92.48Hurt
simple ** marking of highlighted textOriginal prompt · Pronoun P. Score (Qwen3-4B)P. score: 93.14 -> 95.76Improved
apply SEKA to all passages instead of middle only (ablation)SEKA · lost-in-the-middle exact matchexact matchHurt
steer middle passages with SEKASEKA · lost-in-the-middle exact matchexact matchImproved
apply PASTA on lost-in-the-middlePASTA · lost-in-the-middle exact matchexact matchHurt
use SEKA instead of PASTA for inferenceQwen3-8B-Base · inference time per sampleseconds: PASTA +1.03s vs SEKA +0.03sImproved

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