Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Quantifying Point-Prediction Uncertainty in Neural Networks via Residual Estimation with an I/O Kernel
Xin Qiu, Elliot Meyerson, Risto Miikkulainen
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 |
|---|---|---|---|
| apply RIO residual estimation with an I/O kernel to a pretrained NN | standard NN regressors · 12 real-world regression datasets | RMSE | Improved |
| apply RIO to a pretrained DenseNet age estimator | DenseNet · IMDB Age Estimation | MAE: 7.43 -> 6.35 | Improved |
| add residual estimation to direct outcome fitting (ablation) | RIO variants · 12 real-world regression datasets | RMSE | Improved |
| add the I/O kernel rather than using input-only or output-only kernels (ablation) | RIO variants · 12 real-world regression datasets | RMSE | Improved |
| use residual estimation with input kernel only instead of full RIO (ablation) | RIO · yacht | RMSE: 1.46±0.49 -> 2.03±0.73 | Hurt |
| use residual estimation with output kernel only instead of full RIO (ablation) | RIO · yacht | RMSE: 1.46±0.49 -> 1.88±0.66 | Hurt |
| estimate outcomes directly with an output-only kernel instead of full RIO (ablation) | RIO · yacht | RMSE: 1.46±0.49 -> 1.86±0.64 | Hurt |
| estimate outcomes directly with the I/O kernel instead of fitting residuals (ablation) | RIO · yacht | RMSE: 1.46±0.49 -> 1.58±0.52 | Mixed |
| use residual estimation with input kernel only instead of full RIO (ablation) | RIO · ENB/h | RMSE: 0.70±0.38 -> 0.79±0.46 | Hurt |
| use residual estimation with output kernel only instead of full RIO (ablation) | RIO · ENB/h | RMSE: 0.70±0.38 -> 0.80±0.43 | Hurt |
| estimate outcomes directly with output-only kernel instead of full RIO (ablation) | RIO · ENB/h | RMSE: 0.70±0.38 -> 0.88±0.48 | Hurt |
| estimate outcomes directly with the I/O kernel instead of fitting residuals (ablation) | RIO · ENB/h | RMSE: 0.70±0.38 -> 0.76±0.41 | Hurt |
Closest papers by meaning
- NOMU: Neural Optimization-based Model Uncertainty (ICML 2022 · similarity 0.88)
- Uncertainty Quantification with the Empirical Neural Tangent Kernel (NeurIPS 2025 · similarity 0.86)
- Uncertainty Quantification for Deep Regression using Contextualised Normalizing Flows (NeurIPS 2025 · similarity 0.86)
- SURE: SUrvey REcipes for building reliable and robust deep networks (CVPR 2024 · similarity 0.85)
- Approximating Full Conformal Prediction for Neural Network Regression with Gauss-Newton Influence (ICLR 2025 · similarity 0.85)
- Transitional Uncertainty with Layered Intermediate Predictions (ICML 2024 · 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.
