AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Attribution-Guided Decoding

Piotr Komorowski, Elena Golimblevskaia, Reduan Achtibat, Thomas Wiegand, Sebastian Lapuschkin, Wojciech Samek

ICLR 2026 · Large language models › Language, safety and interpretability · approach: Parallel and early-exit decoding

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 LRP attribution instead of Input x Gradient in AGD (ablation)AGD · IHEvalPrompt-level accuracy: 70.8 -> 79.1 (Llama 3.1 8B)Improved
Add entropy-gating to AGD LRP (ablation)AGD LRP · IHEvalQuality score: 73.2 -> 76.4 (Llama 3.1 8B)Improved
Add entropy-gating to AGD LRP (ablation)AGD LRP · IHEvalPrompt-level accuracy: 79.1 -> 74.5 (Llama 3.1 8B)Hurt
Add entropy-gating to AGD LRP (ablation)AGD LRP · Open-book TriviaQA efficiencyCost relative to greedy: 2.20x -> 1.60x (Llama 3.1 8B)Improved
Apply AGD LRP guidance at every stepLlama 3.1 (8B) · IHEvalQuality score: 81.3 -> 73.2Hurt
Use AGD with LRP on system prompt embeddings vs greedy decodingLlama 3.1 (8B) · IHEvalPrompt-level accuracy: 66.0 -> 79.1Improved
Lower the minimum probability threshold to extremely low values (ablation)AGD · IHEvalAdherence and qualityHurt
Raise the minimum probability threshold (ablation)AGD · IHEvalPrompt-level accuracyHurt
Use only 10 heads instead of 100 (ablation)AGD LRP heads · TriviaQA closed-bookRecall: 82.4 -> 81.8Hurt
Use more heads (200 instead of 100) (ablation)AGD LRP heads · TriviaQA closed-bookRecall: 82.4 -> 82.4No clear effect
Use AGD toward context embeddings vs greedy decoding in noisy retrievalLlama 3.1 (8B) · HotPotQA distractorRecall: 81.3 -> 87.9Improved
Guide towards parametric knowledge heads in closed-book QAGemma 3 (4B) · TriviaQA closed-bookRecall: 61.9 -> 61.5No 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