Research map / Label-efficient and robust learning / Domain adaptation and generalization
Learning to Learn with Variational Information Bottleneck for Domain Generalization
Yingjun Du, Jun Xu, Huan Xiong, Qiang Qiu, Xiantong Zhen, Cees G. M. Snoek, Ling Shao
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 |
|---|---|---|---|
| model classifier parameters as distributions (probabilistic baseline) instead of plain fine-tuned AlexNet (ablation) | AlexNet · VLCS (leave-one-domain-out) mean | accuracy: 72.02 -> 73.37 | Improved |
| model classifier parameters as distributions (probabilistic baseline) instead of plain fine-tuned AlexNet (ablation) | AlexNet · PACS (leave-one-domain-out) mean | accuracy: 69.28 -> 72.63 | Improved |
| add regular VIB regularization to probabilistic baseline (ablation) | Baseline · VLCS (leave-one-domain-out) mean | accuracy: 73.37 -> 74.01 | Improved |
| add regular VIB regularization to probabilistic baseline (ablation) | Baseline · PACS (leave-one-domain-out) mean | accuracy: 72.63 -> 73.37 | Improved |
| use meta variational information bottleneck instead of regular VIB (ablation) | Baseline with VIB · PACS (leave-one-domain-out) mean | accuracy: 73.37 -> 75.74 | Improved |
| use meta variational information bottleneck instead of regular VIB (ablation) | Baseline with VIB · VLCS (leave-one-domain-out) mean | accuracy: 74.01 -> 74.54 | Improved |
| set beta to 0.01 rather than 1 in MetaVIB (ablation) | MetaVIB · PACS (leave-one-domain-out) mean | accuracy: 72.02 -> 75.74 | Improved |
| set beta to 0.01 rather than 0.001 or 0.1 (ablation) | MetaVIB · PACS (leave-one-domain-out) mean | accuracy: 73.89 -> 75.74 | Improved |
| set beta to 0.01 rather than 1 on VLCS (ablation) | MetaVIB · VLCS (leave-one-domain-out) mean | accuracy: 71.83 -> 74.54 | Improved |
| set beta to 0.001 rather than 1 on Rotated MNIST (ablation) | MetaVIB · Rotated MNIST (leave-one-angle-out) mean | accuracy: 94.78 -> 97.08 | Improved |
| use 10 Monte Carlo samples instead of 1 (ablation) | MetaVIB · PACS (leave-one-domain-out) mean | accuracy: 72.93 -> 75.74 | Improved |
| increase Monte Carlo samples from 10 to 50 or 100 (ablation) | MetaVIB · PACS (leave-one-domain-out) mean | accuracy: 75.74 -> 75.73 | No clear effect |
Closest papers by meaning
- Learning to Generalize: Meta-Learning for Domain Generalization (AAAI 2018 · similarity 0.90)
- Open Domain Generalization with Domain-Augmented Meta-Learning (CVPR 2021 · similarity 0.90)
- A Bit More Bayesian: Domain-Invariant Learning with Uncertainty (ICML 2021 · similarity 0.90)
- Meta-Causal Learning for Single Domain Generalization (CVPR 2023 · similarity 0.90)
- MetaReg: Towards Domain Generalization using Meta-Regularization (NeurIPS 2018 · similarity 0.89)
- Robust Domain Generalization under Divergent Marginal and Conditional Distributions (NeurIPS 2026 · similarity 0.88)
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.
