AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Q-Delta: Beyond Key–Value Associative State Evolution

Sumin Park, Seojin Kim, Noseong Park

ICML 2026 · Architectures and efficiency › Attention and transformers · approach: State space and recurrent sequence models · 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 query-conditioned feedback to the delta-rule state update (Q-Delta vs Gated DeltaNet)Gated DeltaNet · Zero-shot commonsense avg, 340Maverage accuracy: 46.01 -> 47.24Improved
add query-conditioned feedback (Q-Delta vs Gated DeltaNet), 1.3BGated DeltaNet · Zero-shot commonsense avg, 1.3Baverage accuracy: 52.77 -> 53.47Improved
add query feedback for long-context retrievalGated DeltaNet · S-NIAH average, 1.3Baccuracy: 83.51 -> 90.02Improved
add query feedback for LAMBADA perplexity at 340MGated DeltaNet · LAMBADA, 340Mperplexity: 36.27 -> 32.67Improved
compare with Mamba2 on perplexity at 340MQ-Delta · WikiText/LAMBADA, 340Mperplexity: Mamba2 26.60 / 30.35 vs Q-Delta 26.89 / 32.67Hurt
add query feedback on SQuAD recall taskGated DeltaNet · Recall-intensive SQuAD, 340Maccuracy: 56.0 -> 52.3Hurt
add query feedback on SWDE and FDA recall tasksGated DeltaNet · Recall-intensive SWDE, FDA, 340Maccuracy: SWDE 29.1 -> 31.6; FDA 18.9 -> 22.0Improved
learn lambda_t end-to-end instead of fixed scalar 0.5 (ablation)Q-Delta · Zero-shot commonsense avg, 340Maverage accuracy: 47.20 -> 47.24Improved
use learnable lambda_t instead of fixed lambda = 0.8 (ablation)Q-Delta · Zero-shot commonsense avg, 340Maverage accuracy: 46.42 -> 47.24Improved
add adaptive gating of query feedback (vs lambda_t = 1) (ablation)Q-Delta · LAMBADA ppl, 340Mperplexity: 35.21 -> 32.67Improved
add state decay gate alpha_t (ablation)Q-Delta · Zero-shot commonsense avg, 340Maverage accuracy: 45.86 -> 47.24Improved
fixed lambda on WikiText perplexity vs learnable (ablation)Q-Delta · WikiText, 340Mperplexity: 26.89 vs 26.52-26.96 across variantsNo 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