Research map / Label-efficient and robust learning / Recognition and long-tail
Variational Distillation for Multi-View Learning
Xudong Tian, Zhizhong Zhang, Cong Wang, Wensheng Zhang, Yanyun Qu, Lizhuang Ma, Zongze Wu, Yuan Xie, Dacheng Tao
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 conventional IB with VSD (single view) (ablation) | IB re-ID model · SYSU-MM01 all-search | Rank-1 and mAP: 27.68%@Rank-1 and 24.58%@mAP | Improved |
| replace conventional IB with VCD and VMD (cross view) (ablation) | IB re-ID model · SYSU-MM01 all-search | Rank-1 and mAP: 23.14%@Rank-1 and 22.82%@mAP | Improved |
| add conventional IB training to shared encoder (ablation) | E_S encoder · SYSU-MM01 all-search | Rank-1: 53.19 -> 38.02 | Hurt |
| add information bottleneck architecture (ablation) | E_S encoder · SYSU-MM01 all-search | Rank-1: 53.19 -> 56.95 | Improved |
| add VCD and VMD to IB architecture (ablation) | E_S+B_S · SYSU-MM01 all-search | Rank-1: 56.95 -> 61.16 | Improved |
| add VSD to IB architecture (ablation) | E_I/V+B_I/V · SYSU-MM01 all-search | Rank-1: 62.26 -> 69.33 | Improved |
| add variational distillation to three-branch IB model (ablation) | E_S/I/V+B_S/I/V · SYSU-MM01 all-search | Rank-1: 64.15 -> 70.02 | Improved |
| add third (uniform image) view to 2-view MV2D | MV2D re-ID · SYSU-MM01 all-search | Rank-1: 70.02 -> 71.65 | Improved |
| add 3-view MV2D to baseline | MV2D re-ID · RegDB Visible2Thermal | Rank-1: 79.9 -> 83.1 | Improved |
| add MV2D to multi-view baseline (ablation) | MvNNcor-based model · Caltech-101 | accuracy: 82.66 -> 85.93 | Improved |
| add MV2D to multi-view baseline (ablation) | MvNNcor-based model · Hand | accuracy: 99.29 -> 99.40 | No clear effect |
| add conventional IB training to encoder (ablation) | E_theta · Caltech-101 | accuracy: 81.28 -> 64.29 | Hurt |
Closest papers by meaning
- Differentiable Information Bottleneck for Deterministic Multi-view Clustering (CVPR 2024 · similarity 0.89)
- Rethinking Multi-view Representation Learning via Distilled Disentangling (CVPR 2024 · similarity 0.89)
- Multi-View Causal Representation Learning with Partial Observability (ICLR 2024 · similarity 0.86)
- Aligning Multimodal Representations through an Information Bottleneck (ICML 2025 · similarity 0.86)
- Partial Multi-View Multi-Label Classification via Semantic Invariance Learning and Prototype Modeling (ICML 2024 · similarity 0.86)
- Incomplete Multi-View Multi-label Learning via Disentangled Representation and Label Semantic Embedding (CVPR 2025 · similarity 0.85)
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.
