AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Query Lens: Interpreting Sparse Key-Value Features with Indirect Effects

Hwiyeong Lee, Ingyu Bang, Uiji Hwang, Hyelim Lim, Taeuk Kim

ICML 2026 · Architectures and efficiency › RNNs and normalization · approach: Sparse dictionary 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
add indirect (module-mediated) effects via full stream transition on key side (ablation)Logit Lens key variant · GPT-2 Small input scoreinput score (%): 7.84 -> 39.32Improved
add indirect effects on value side (ablation)Logit Lens value variant · GPT-2 Small output scoreoutput score (%): 12.57 -> 15.24Improved
add indirect effects on value side (ablation)Logit Lens value variant · Gemma-3-270M output scoreoutput score (%): 6.13 -> 10.63Improved
use key-side QL for output score (ablation)QL-Key · GPT-2 Small output scoreoutput score (%): 15.24 -> 1.97Hurt
use Query Lens instead of Tuned LensTuned Lens · GPT-2 Small input scoreinput score (%): 8.24 -> 39.32Improved
use Query Lens instead of Token Change clampingTC · GPT-2 Small input scoreinput score (%): 31.47 -> 39.32Improved
use Query Lens token sets instead of Logit LensLL · Qwen-3-1.7B LLM-judged interpretabilityinterpretability score: 5.06 -> 6.96Improved
use first-order transition instead of full transition (ablation)Query Lens · Gemma-3-1B output score layer 22 (near)output score: 0.156 -> 0.176Improved
use full transition instead of Logit Lens near output endpoint (ablation)Query Lens · Gemma-3-1B layer 22output score: 0.183 -> 0.156Hurt
use full transition instead of Logit Lens far from endpoint (ablation)Query Lens · Gemma-3-1B layer 7output score: 0.010 -> 0.046Improved
increase top-k from 5 to 25 (ablation)Query Lens · GPT-2 Small input scoreinput score (%): 30.37 -> 39.32Improved
QL vs LL at k=5 for output score (ablation)Query Lens · Qwen-3-1.7Boutput score: 6.50 vs 6.65No clear effect

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