Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Variational Multi-Task Learning with Gumbel-Softmax Priors
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use variational inference for both representations and classifiers (ablation) | VMTL · Office-Home 5% | average accuracy: 50.4{ $\pm$0.1} -> 58.3{ $\pm$0.1} | Helped |
| Add variational inference for representations (ablation) | VMTL · Office-Home 5% | average accuracy: 50.4{ $\pm$0.1} -> 57.4{ $\pm$0.0} | Helped |
| Add variational inference for classifiers (ablation) | VMTL · Office-Home 5% | average accuracy: 50.4{ $\pm$0.1} -> 56.8{ $\pm$0.1} | Helped |
| Use Gumbel-Softmax priors instead of mean priors (ablation) | VMTL · Office-Home 5% | average accuracy: 57.5{ $\pm$0.0} -> 58.3{ $\pm$0.1} | Helped |
| Use Gumbel-Softmax priors instead of learnable weighted priors (ablation) | VMTL · Office-Home 5% | average accuracy: 57.2{ $\pm$0.2} -> 58.3{ $\pm$0.1} | Helped |
| Use VMTL rather than Long et al. | Office-Home · Office-Home 5% | average accuracy: 57.4{ $\pm$0.1} -> 58.3{ $\pm$0.1} | Helped |
| Use VMTL rather than Bakker et al. | Office-Caltech · Office-Caltech 10% | average accuracy: 95.0{ $\pm$0.2} -> 95.3{ $\pm$0.0} | Helped |
| Use VMTL rather than Kendall et al. | ImageCLEF · ImageCLEF 5% | average accuracy: 74.6{ $\pm$0.2} -> 76.2{ $\pm$0.3} | Helped |
| Use VMTL rather than Qian et al. | DomainNet · DomainNet 2% | average accuracy: 29.0{ $\pm$0.1} -> 29.8{ $\pm$0.0} | Helped |
Closest papers by meaning
- Variational Task Vector Composition (NeurIPS 2025 · similarity 0.90)
- Bayesian Uncertainty for Gradient Aggregation in Multi-Task Learning (ICML 2024 · similarity 0.88)
- Identifying and Mitigating Spurious Correlation in Multi-Task Learning (CVPR 2025 · similarity 0.87)
- Towards Consistent Multi-Task Learning: Unlocking the Potential of Task-Specific Parameters (CVPR 2025 · similarity 0.86)
- Distributional Alignment as a Criterion for Designing Task Vectors in In-Context Learning (NeurIPS 2026 · similarity 0.86)
- NTKMTL: Mitigating Task Imbalance in Multi-Task Learning from Neural Tangent Kernel Perspective (NeurIPS 2025 · 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.