AI Paper Atlas

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

Mechanism of Task-oriented Information Removal in In-context Learning

Hakaze Cho, Haolin Yang, Gouki Minegishi, Naoya Inoue

ICLR 2026 · Large language models › Language, safety and interpretability · approach: In-context learning analysis · 1 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 denoising heads (not ablated) (ablation)8-shot ICL, Llama 3.2-1B · Avg 6 datasets, unseen labelaccuracy (%): 14.54 -> 1.06Improved
add denoising heads (not ablated) (ablation)8-shot ICL, Llama 3-8B · Avg 6 datasets, unseen labelaccuracy (%): 21.66 -> 7.52Improved
add denoising heads (not ablated) (ablation)8-shot ICL, Qwen 2.5-3B · Avg 6 datasets, unseen labelaccuracy (%): 23.49 -> 11.59Improved
add denoising heads (not ablated) (ablation)8-shot ICL, Qwen 2.5-3B Instruct · Avg 6 datasets, unseen labelaccuracy (%): 48.70 -> 37.51Improved
add denoising heads (not ablated) (ablation)8-shot ICL, Llama 3-13B Instruct · Avg SST-2, unseen labelaccuracy (%): 29.59 -> 10.35Improved
add denoising heads (not ablated) (ablation)8-shot ICL, Llama 3.2-1B · Avg 6 datasets, random-sample demosaccuracy (%): 71.02 -> 55.04Improved
add denoising heads (not ablated) (ablation)8-shot ICL, Llama 3.2-1B · Avg 6 datasets, seen labelaccuracy (%): 73.40 -> 56.52Improved
add denoising heads (not ablated) (ablation)8-shot ICL, Llama 3-8B · Avg 6 datasets, random-sample demosaccuracy (%): 77.63 -> 72.05Improved
add denoising heads (not ablated) (ablation)8-shot ICL, Qwen 2.5-3B Instruct · Avg 6 datasets, random-sample demosaccuracy (%): 77.24 -> 75.57No clear effect
ablate random heads instead of DHs (ablation)8-shot ICL, Qwen 2.5-3B · Avg 6 datasets, random-sample demosaccuracy (%): 73.97 -> 74.31No clear effect
fine-tune only W_dec (freeze W_enc) to transfer verbalization to A/B (ablation)SST-2 filter, Llama 3.2-1B · SST-2accuracy (%): 0.00 -> 0.84Improved
train filter on both parts versus fine-tuning only W_enc (ablation)SST-2 filter, Llama 3.2-1B · SST-2accuracy (%): 0.00 -> 0.88Improved

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