Research map / Label-efficient and robust learning / Semi-supervised learning
Semi-Supervised Deep Regression with Uncertainty Consistency and Variational Model Ensembling via Bayesian Neural Networks
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 UCVME instead of the strongest reported alternative | Semi-supervised age estimation with ResNet50 · UTKFace, 5%, 10%, and 20% labeled | MAE: 5.97 $\pm$ 0.06 -> 5.84 $\pm$ 0.06; 5.39 $\pm$ 0.04 -> 5.26 $\pm$ 0.02; 4.97 $\pm$ 0.03 -> 4.85 $\pm$ 0.03 | Helped |
| Use UCVME instead of the strongest reported alternative | Semi-supervised age estimation with ResNet50 · UTKFace, 5%, 10%, and 20% labeled | R²: 47.4% $\pm$ 1.0 -> 49.4% $\pm$ 0.7; 56.6% $\pm$ 0.4 -> 57.9% $\pm$ 0.3; 62.7% $\pm$ 0.2 -> 64.3% $\pm$ 0.5 | Helped |
| Use UCVME instead of the strongest reported alternative | Semi-supervised ejection fraction estimation with R2+1D · EchoNet-Dynamic, 1/16, 1/8, and 1/4 labeled | MAE: 5.90 $\pm$ 0.11 -> 5.77 $\pm$ 0.04; 5.31 $\pm$ 0.02 -> 5.10 $\pm$ 0.05; 4.57 $\pm$ 0.02 -> 4.37 $\pm$ 0.05 | Helped |
| Use UCVME instead of the strongest reported alternative | Semi-supervised ejection fraction estimation with R2+1D · EchoNet-Dynamic, 1/16, 1/8, and 1/4 labeled | R²: 56.3% $\pm$ 1.0 -> 57.8% $\pm$ 0.6; 64.9% $\pm$ 0.3 -> 66.6% $\pm$ 0.5; 74.5% $\pm$ 0.1 -> 76.3% $\pm$ 0.6 | Helped |
| Add aleatoric uncertainty consistency loss (ablation) | Baseline co-trained BNNs · UTKFace, 10% labeled | MAE: 5.40 $\pm$ 0.03 -> 5.30 $\pm$ 0.01 | Helped |
| Add variational model ensembling (ablation) | Baseline co-trained BNNs · UTKFace, 10% labeled | MAE: 5.40 $\pm$ 0.03 -> 5.31 $\pm$ 0.02 | Helped |
| Add uncertainty consistency loss and variational model ensembling (ablation) | Baseline co-trained BNNs · UTKFace, 10% labeled | MAE: 5.40 $\pm$ 0.03 -> 5.26 $\pm$ 0.02 | Helped |
| Add uncertainty consistency loss and variational model ensembling (ablation) | Baseline co-trained BNNs · UTKFace, 10% labeled | R²: 56.3% $\pm$ 0.5 -> 57.9% $\pm$ 0.3 | Helped |
| Set variational ensemble sampling to T=5 instead of T=1 (ablation) | UCVME · Age estimation computation | computation cost (G): 21 -> 49 | Hurt |
Closest papers by meaning
- Effective Bayesian Heteroscedastic Regression with Deep Neural Networks (NeurIPS 2023 · similarity 0.88)
- Semi-Supervised Regression with Heteroscedastic Pseudo-Labels (NeurIPS 2025 · similarity 0.87)
- SURE: SUrvey REcipes for building reliable and robust deep networks (CVPR 2024 · similarity 0.86)
- Efficient Analytic Uncertainty Quantification for Multimodal Regression (NeurIPS 2026 · similarity 0.86)
- VCT: Training Consistency Models with Variational Noise Coupling (ICML 2025 · similarity 0.85)
- Towards Self-Supervised Covariance Estimation in Deep Heteroscedastic Regression (ICLR 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.