AI Paper Atlas

Research map / Learning theory and algorithms / Prediction and decision losses

Memorization Through the Lens of Sample Gradients

Deepak Ravikumar, Efstathia Soufleri, Abolfazl Hashemi, Kaushik Roy

ICLR 2026 · Learning theory and algorithms › Prediction and decision losses · approach: Loss landscape analysis

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
stop training at the peak of weight norm / sample gradient instead of last epochResNet18 on CIFAR-100 · CIFAR-100 (LiRA attack)MIA AUROC: 85.48 -> 55.98Improved
stop training at the peak of weight norm / sample gradient instead of last epochResNet18 on CIFAR-100 · CIFAR-100 (curvature attack)MIA AUROC: 85.52 -> 59.42Improved
stop training at the peak of weight norm / sample gradient instead of last epochResNet18 on CIFAR-100 · CIFAR-100UCE: 2.1839 -> 2.0821Improved
stop training at the peak of weight norm / sample gradient instead of last epochResNet18 on CIFAR-100 · CIFAR-100ECE: 0.1017 -> 0.1382Hurt
stop training at the peak of weight norm / sample gradient instead of last epochResNet18 on CIFAR-100 · CIFAR-100accuracy: 0.7492 -> 0.6306Hurt
replace validation-loss early stopping with sample gradient early stoppingResNet18 on CIFAR-100 · CIFAR-100accuracy: 0.6262 -> 0.6306No clear effect
stop training early and accumulate loss (SGAL) instead of full-training CSGCSG · CIFAR-100 (Inception)cosine similarity with memorization: 0.84 -> 0.86Improved
stop training early and accumulate loss (SGAL) instead of full-training CSGCSG · ImageNet (ResNet50)Pearson correlation with memorization: 0.52 -> 0.62Improved
use SGAL instead of CSG for mislabel detectionCSG · CIFAR-10 mislabel detection (30% noise)AUROC: 0.9935 -> 0.8470Hurt
train ResNet18 instead of Inception for CSG (ablation)CSG · CIFAR-100cosine similarity with memorization: 0.84 -> 0.77Hurt
use ResNet18 instead of ResNet50 for CSG (ablation)CSG · ImageNetcosine similarity with memorization: 0.78 -> 0.68Hurt
accumulate input gradient over training instead of using last-5-checkpoint variance (VoG)VoG · CIFAR-100 (Inception)cosine similarity with memorization: 0.33 -> 0.84Improved

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