Research map / Vision-language and multimodal / Cross-modal learning
Causal Disentanglement and Cross-Modal Alignment for Enhanced Few-Shot Learning
Tianjiao Jiang, Zhen Zhang, Yuhang Liu, Javen Qinfeng Shi
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 |
|---|---|---|---|
| add ICA disentanglement of cache keys (ablation) | CCA-FT · 11-dataset average, 1-shot | accuracy: 64.94 -> 66.00 | Improved |
| add ICA disentanglement of cache keys (ablation) | CCA-FT · 11-dataset average, 16-shot | accuracy: 75.95 -> 77.60 | Improved |
| fine-tune the cache adapter (ablation) | CCA-FT · 11-dataset average, 16-shot | accuracy: 75.75 -> 77.60 | Improved |
| fine-tune the CLIP text classifier (ablation) | CCA-FT · 11-dataset average, 16-shot | accuracy: 77.00 -> 77.60 | Improved |
| add cross-attention fused features (ablation) | CCA-FT · 11-dataset average, 16-shot | accuracy: 77.43 -> 77.60 | Improved |
| add cross-attention fused features (ablation) | CCA-FT · 11-dataset average, 1-shot | accuracy: 65.81 -> 66.00 | Improved |
| fine-tune CCA (CCA-FT) instead of training-free CCA (ablation) | CCA · ImageNet 16-shot | accuracy: 63.00 -> 66.04 | Improved |
| replace Tip-Adapter by training-free CCA | Tip-Adapter · ImageNet-Sketch | accuracy: 35.71 -> 36.60 | Improved |
| replace Tip-Adapter-F by CCA-FT under Gaussian noise | Tip-Adapter-F · Gaussian noise, 16-shot average | accuracy: 60.04 -> 62.48 | Improved |
| replace Tip-Adapter-F by CCA-FT under FGSM attack | Tip-Adapter-F · FGSM, 16-shot average | accuracy: 33.21 -> 35.48 | Improved |
| replace Tip-Adapter-F by CCA-FT on ImageNet-Sketch | Tip-Adapter-F · ImageNet-Sketch | accuracy: 35.00 -> 35.87 | Improved |
| train CCA-FT 20 epochs instead of CoOp 200 epochs | CoOp · ImageNet 16-shot | accuracy: 7.5h -> 4.9min; 62.95 -> 66.04 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Tip-Adapter — Tip-Adapter: Training-Free Adaption of CLIP for Few-Shot Classification (ECCV 2022) · better on 5 of 5 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Post-pre-training for Modality Alignment in Vision-Language Foundation Models (CVPR 2025 · similarity 0.87)
- Understanding Transferable Representation Learning and Zero-shot Transfer in CLIP (ICLR 2024 · similarity 0.87)
- SmartCLIP: Modular Vision-language Alignment with Identification Guarantees (CVPR 2025 · similarity 0.87)
- Cross the Gap: Exposing the Intra-modal Misalignment in CLIP via Modality Inversion (ICLR 2025 · similarity 0.87)
- On the Value of Cross-Modal Misalignment in Multimodal Representation Learning (NeurIPS 2025 · similarity 0.87)
- RePercENT: Scaling Disentangled Representation Learning Beyond Two Modalities (NeurIPS 2026 · similarity 0.86)
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.
