Research map / Vision-language and multimodal / CLIP and zero-shot
UniCon: Unified Framework for Efficient Contrastive Alignment via Kernels
Hangke Sui, Yuqing Wang, Minh N. Do
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 changed | Where | Measured | Effect |
|---|---|---|---|
| replace SGD-CLIP with UniCon spectral update | RN-50 + SBERT · Flickr30k retrieval | average R@10: .219 -> .515 | Improved |
| replace SGD-CLIP with UniCon spectral update | CLIP ViT-B/32 · Flickr30k retrieval | average R@10: .597 -> .701 | Improved |
| replace SGD-CLIP with UniCon spectral update | RN-18 + SBERT · Flickr30k retrieval | image-to-text R@10: .221 -> .145 | Hurt |
| replace SGD-CLIP with UniCon spectral update | RN-18 + SBERT · Flickr30k retrieval | train time: 45.6 s -> 1.7 s | Improved |
| replace SGD-CLIP with UniCon spectral update | CLIP ViT-B/32 · MSCOCO retrieval | image-to-text R@1: .128 -> .329 | Improved |
| replace SGD-CLIP with UniCon spectral update | CLIP ViT-B/32 · MSCOCO retrieval | train time (s): 1066.60 -> 11.15 | Improved |
| replace SGD-CLIP with UniCon spectral update | RN-50 + SBERT · MSCOCO retrieval | train time (s): 5121.72 -> 11.11 | Improved |
| replace SGD-CLIP with UniCon spectral update | frozen ResNet-18 · CIFAR-10 linear probe | accuracy (%): 62.21 -> 61.82 | No clear effect |
| replace SGD-CLIP with UniCon spectral update | frozen ResNet-18 · CIFAR-10 linear probe | train time: 41.98s -> 23.38s | Improved |
| replace SGD-CLIP with UniCon spectral update | MLP encoder · Synthetic nonlinear latent-factor | matching accuracy (%): 84 -> 86 | Improved |
Closest papers by meaning
- Seeing the Image: Prioritizing Visual Correlation by Contrastive Alignment (NeurIPS 2024 · similarity 0.88)
- UNIALIGN: Scaling Multimodal Alignment within One Unified Model (CVPR 2025 · similarity 0.87)
- With Limited Data for Multimodal Alignment, Let the STRUCTURE Guide You (NeurIPS 2025 · similarity 0.87)
- CovMatch: Cross-Covariance Guided Multimodal Dataset Distillation with Trainable Text Encoder (NeurIPS 2025 · similarity 0.87)
- UniGen: Enhanced Training & Test-Time Strategies for Unified Multimodal Understanding and Generation (NeurIPS 2025 · similarity 0.87)
- Extending Multi-modal Contrastive Representations (NeurIPS 2024 · similarity 0.87)
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.
