AI Paper Atlas

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

NeuCLIP: Efficient Large-Scale CLIP Training with Neural Normalizer Optimization

Xiyuan Wei, Chih-Jen Lin, Tianbao Yang

ICLR 2026 · Label-efficient and robust learning › Recognition and long-tail · approach: Multilingual language representation pretraining

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
replace OpenCLIP/FastCLIP with NeuCLIPCLIP training · DFN-14MDatacomp Average: 37.78 -> 39.16 (OpenCLIP), 38.45 -> 39.16 (FastCLIP)Improved
replace OpenCLIP with NeuCLIP at largest scaleCLIP training · DFN-1BDatacomp Average: 56.25 -> 57.34Improved
use a unified objective instead of separate objectives for CLIP and NPN (ablation)NeuCLIP · DFN-14MDatacomp Average: 38.63 -> 39.16Improved
use single-layer inductive-bias NPN instead of MLP (ablation)NeuCLIP · DFN-14MDatacomp Average: 38.58 -> 39.16Improved
re-initialize NPNs every 500 iterations instead of never (ablation)NeuCLIP · DFN-14MDatacomp Average: 38.48 -> 39.16Improved
restart too rarely (Tr 500 -> 2,500) (ablation)NeuCLIP · CC12MDatacomp Average: 31.89 -> 31.47Hurt
restart too often (Tr 500 -> 20) (ablation)NeuCLIP · DFN-14MDatacomp Average: 39.16 -> 38.41Hurt
increase NPN updates per iteration from 1 to 10 (ablation)NeuCLIP · DFN-14MDatacomp Average: 39.02 -> 39.16Improved
increase NPN updates beyond 10 (10 -> 50) (ablation)NeuCLIP · DFN-14MDatacomp Average: 39.16 -> 38.03Hurt
increase number of prototypes m from 4,096 to 8,192 (ablation)NeuCLIP · DFN-14MDatacomp Average: 39.16 -> 39.25No clear effect
increase number of prototypes m from 1,024 to 4,096 (ablation)NeuCLIP · DFN-14MDatacomp Average: 38.57 -> 39.16Improved
add NPN training to OpenCLIP-style trainingNeuCLIP · ViT-B/16time overhead: 5.98% overheadNo 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