AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Neural Collapse by Design: Learning Class Prototypes on the Hypersphere

Panagiotis Koromilas, Theodoros Giannakopoulos, Mihalis Nicolaou, Yannis Panagakis

ICML 2026 · Label-efficient and robust learning › Recognition and long-tail · approach: Loss design and classifier theory · 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
normalize features and weights to the hypersphere (NormFace) instead of CE (ablation)classifier learning · CIFAR-10/100, ImageNet-100/1Ktop-1 accuracy: 75.4 -> 76.4 (ImageNet-1K)Improved
use all batch samples as negatives (NTCE) instead of K class prototypes (ablation)NormFace · ImageNet-1Ktop-1 accuracy: 76.4 -> 76.7Improved
exclude same-class instances from the denominator (NONL) (ablation)NTCE · CIFAR-100top-1 accuracy: 72.9 -> 73.6Improved
NONL instead of NTCE on ImageNet-1K (ablation)NTCE/NONL · ImageNet-1Ktop-1 accuracy: 76.7 vs 76.5No clear effect
replace linear probing with fixed class-mean prototypesSCL · CIFAR-10, CIFAR-100, ImageNet-1Ktop-1 accuracy: 75.1 -> 75.1 (ImageNet-1K)No clear effect
replace linear probing with fixed class-mean prototypesSCL · ImageNet-100top-1 accuracy: 84.8 -> 86.8Improved
replace linear probing with normalized linear probing (ablation)SCL · CIFAR-10, CIFAR-100, ImageNettop-1 accuracy: 73.9 -> 73.6 (CIFAR-100)No clear effect
use NONL/NTCE instead of CEclassifier learning · CIFAR-10/100 trainingNC effective rank and alignment: Inter ER 8.6/57.1 -> 9.0/99.0Improved
train SCL with linear probing vs fixed prototypes (ablation)SCL · CIFAR-10/100NC3 weight-feature alignment: 0.99/1.03 -> 0.00/0.00Improved
use NONL instead of CE for pretrainingResNet50 · Transfer learning (8 datasets)mean accuracy: 67.0 -> 70.7Improved
use ETF + DR classifier (ablation)ResNet50 · Transfer learning (8 datasets)mean accuracy: 67.0 -> 54.6Hurt
use NONL instead of CEResNet18 · CIFAR-100-LT (ratio 0.01)accuracy: 37.4 -> 40.0Improved

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