Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Detecting Misclassification Errors in Neural Networks with a Gaussian Process Model
Xin Qiu, 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 |
|---|---|---|---|
| Use RED instead of MCP confidence | Standard NN classifier · 125 UCI datasets | AP-Error mean rank: 2.93± 0.89 -> 1.39±0.61 | Improved |
| Use RED instead of MCP confidence | Standard NN classifier · 125 UCI datasets | AUPR-Error mean rank: 3.06±0.92 -> 1.49±0.78 | Improved |
| Use RED instead of Trust Score | VGG16 · CIFAR-10 | AP-Error (%): 48.76±2.28 -> 49.88±1.99 | Improved |
| Use RED instead of Trust Score | WRN-10-4 · CIFAR-10 | AP-Error (%): 51.15±3.29 -> 52.51±2.81 | Improved |
| Use RED instead of Trust Score | VGG19 · CIFAR-100 | AP-Error (%): 72.63±1.20 -> 73.40±1.05 | Improved |
| Replace the GP with Bayesian linear regression (ablation) | RED · 125 UCI datasets | AP-Error pairwise comparison | Hurt |
| Use RED-variance rather than MCP | OOD detection · SVHN OOD vs CIFAR-10, VGG16 | AP-OOD (%): 82.964±1.850 -> 86.282±2.212 | Improved |
| Use RED-variance rather than RED-mean | RED · 125 UCI datasets with OOD and adversarial samples | AP-OOD and AP-Adversarial | Improved |
| Apply RED to the BNN or MC-Dropout classifier rather than using its internal confidence score | BNN and MC-Dropout classifiers · 125 UCI datasets | AP-Error pairwise comparisons | Improved |
| Use RED rather than MCP when errors are few or classifier accuracy is high | Standard NN classifier · 125 UCI datasets, few errors or high accuracy | AP-Error | Improved |
Closest papers by meaning
- Revisiting Confidence Estimation: Towards Reliable Failure Prediction (TPAMI 2023 · similarity 0.87)
- A Call to Reflect on Evaluation Practices for Failure Detection in Image Classification (ICLR 2023 · similarity 0.86)
- OpenMix: Exploring Outlier Samples for Misclassification Detection (CVPR 2023 · similarity 0.85)
- Generalized Resubstitution for Classification Error Estimation (JMLR 2022 · similarity 0.84)
- Interpretable Failure Detection with Human-Level Concepts (AAAI 2025 · similarity 0.84)
- A Data-Driven Measure of Relative Uncertainty for Misclassification Detection (ICLR 2024 · similarity 0.84)
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.
