AI Paper Atlas

Research map / Large language models / Retrieval and knowledge

Cram Less to Fit More: Training Data Pruning Improves Memorization of Facts

Jiayuan Ye, Vitaly Feldman, Kunal Talwar

ICML 2026 · Large language models › Retrieval and knowledge · approach: Knowledge editing in LLMs · 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 loss-based head selection (LossH-Wiki, alpha 0.2)110m model full-data training · Wikipedia test facts (110m)fact accuracy (accurate fact count): 696.11 -> 929.25Improved
prune facts with loss-based selection instead of random pruning (alpha 0.2) (ablation)random fact pruning · Wikipedia test facts (110m)fact accuracy (accurate fact count): 647.33 -> 929.25Improved
prune facts with loss-based selection instead of random pruning (alpha 0.9) (ablation)random fact pruning · Wikipedia test facts (110m)fact accuracy (accurate fact count): 690.86 -> 733.23Improved
random fact pruning instead of full data (ablation)110m model full-data training · Wikipedia test facts (110m)fact accuracy (accurate fact count): 696.11 -> 690.86No clear effect
loss-based selection at small alpha 0.1 vs full data110m model · Wikipedia test facts (110m)fact accuracy (accurate fact count): 696.11 -> 835.81Improved
loss-based selection on NLU accuracy at alpha 0.2110m model · Wikipedia NLUNLU accuracy: 0.3636 -> 0.3604No clear effect
prune training data to alpha 0.2 vs training a 10x larger model on full data110m model · Wikipedia test facts (110m)fact accuracyImproved
prune data (alpha 0.5) on Knowledge-MMLU110m model · Knowledge-MMLUaccuracyMixed
use fact-level selection instead of token-level selection (ablation)LossH/LossHF · synthetic phonebookfact accuracyImproved
use sum of loss instead of average loss over fact tokens (ablation)LossH/LossHF · synthetic phonebookrank correlation to fact weight-to-bits ratioImproved
use online percentile threshold per batch (ablation)LossH/LossHF · synthetic phonebookcorrelation between loss and fact frequencyImproved
add flattening to head selection (LossHF) (ablation)LossH · power-law phonebookconvergence speedMixed

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