AI Paper Atlas

Research map / Vision-language and multimodal / Cross-modal learning

The Convergent Representation of Contrastive Vision-Language Models: Geometry, Modality Gap and Shared Space Alignment

Lingjie Yi, Raphael Douady, Chao Chen

ICML 2026 · Vision-language and multimodal › Cross-modal learning · approach: Multimodal fusion with missing modalities

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 the SSA shared-space alignment lossCLIP ViT-B/16 (CC12M) · CIFAR-10 zero-shotTop-1 accuracy: 61.6 -> 71.3Improved
add the SSA shared-space alignment lossCLIP ViT-B/16 (CC12M) · CIFAR-100 zero-shotTop-1 accuracy: 28.1 -> 37.8Improved
add the SSA shared-space alignment lossCLIP ViT-B/16 (CC12M) · ImageNet-1K zero-shotTop-1 accuracy: 31.4 -> 33.3Improved
add the SSA loss instead of modality-gap reduction by AlignCLIPCLIP ViT-B/16 (CC12M) · CIFAR-10 zero-shotTop-1 accuracy: 69.4 -> 71.3Improved
add the SSA shared-space alignment lossCLIP ViT-B/16 (CC12M) · MSCOCO image-to-text retrievalR@1: 31.4 -> 34.5Improved
add the SSA shared-space alignment lossCLIP ViT-B/16 (CC12M) · MSCOCO text-to-image retrievalR@1: 20.5 -> 22.8Improved
share encoder parameters to reduce modality gap (SharedCLIP)CLIP ViT-B/16 (CC12M) · CIFAR-10 zero-shotTop-1 accuracy: 61.6 -> 56.9Hurt
post-hoc project embeddings onto the shared space to shrink the modality gapCLIP ViT-B/32 · MSCOCO image-to-text retrievalR@1: 48.40 -> 39.26Hurt
post-hoc project embeddings onto the shared space to shrink the modality gapCLIP ViT-B/32 · MSCOCO text-to-image retrievalR@1: 29.89 -> 27.26Hurt
post-hoc normalize the shared-space projectionsCLIP ViT-B/32 · MSCOCO image-to-text retrievalR@1: 48.40 -> 36.70Hurt
larger shared space alignment across 78 VLMs78 pretrained VLMs · MSCOCOT2I R@1 correlation: Kendall tau 0.733Improved
larger cone concentration kappa_avg across 78 VLMs78 pretrained VLMs · MSCOCOmodality gap correlation: Kendall tau 0.047No 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