Research map / Vision-language and multimodal / Cross-modal learning
Calibrated Information Bottleneck for Trusted Multi-modal Clustering
Shizhe Hu, Zhangwen Gou, Shuaiju Li, Jin Qin, Xiaoheng Jiang, Pei Lv, Mingliang Xu
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 calibration loss LcaliH on top of LIB + LcluH (ablation) | CLIB · Caltech-3V | ACC (%): 60.6 -> 67.8 | Improved |
| add LcaliH on top of LIB + LcluH (ablation) | CLIB · Caltech-3V | ECE (%): 25.5 -> 11.6 | Improved |
| add reconstruction loss Lre (ablation) | CLIB · Caltech-3V | ACC (%): 60.6 -> 64.6 | Improved |
| add consistency loss Lcon (ablation) | CLIB · Caltech-3V | ECE (%): 25.5 -> 12.4 | Improved |
| add Lcon (ablation) | CLIB · Caltech-3V | ACC (%): 60.6 -> 61.1 | No clear effect |
| add Lcon (ablation) | CLIB · IAPR | ACC (%): 39.1 -> 37.8 | Hurt |
| add warm-up stage for IB before calibration (ablation) | CLIB · Caltech-3V | ACC (%): 61.7 -> 77.8 | Improved |
| stop gradient from calibration heads to IB (ablation) | CLIB · Caltech-3V | ACC (%): 65.9 -> 77.8 | Improved |
| use dynamic quality-score screening (S2) instead of max-probability selection (ablation) | CLIB · Caltech-3V | ACC (%): 70.6 -> 77.8 | Improved |
| use screening strategy S2 instead of S1 (ablation) | CLIB · Caltech-3V | ACC (%): 65.6 -> 77.8 | Improved |
| use screening strategy S2 instead of S3 (ablation) | CLIB · MIRFlickr | ACC (%): 43.1 -> 55.4 | Improved |
| continue calibration training from epoch 101 to 200 on difficult samples | CLIB · Caltech-3V (hard 20% samples) | ACC (%): 18.9 -> 46.7 | Improved |
Closest papers by meaning
- A Peer-review Look on Multi-modal Clustering: An Information Bottleneck Realization Method (ICML 2025 · similarity 0.91)
- Super Deep Contrastive Information Bottleneck for Multi-modal Clustering (ICML 2025 · similarity 0.89)
- Multi-aspect Self-guided Deep Information Bottleneck for Multi-modal Clustering (AAAI 2025 · similarity 0.89)
- Towards Calibrated Deep Clustering Network (ICLR 2025 · similarity 0.86)
- Differentiable Information Bottleneck for Deterministic Multi-view Clustering (CVPR 2024 · similarity 0.86)
- Weighted Conformal Clustering (NeurIPS 2026 · similarity 0.84)
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.
