AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Combining Human Predictions with Model Probabilities via Confusion Matrices and Calibration

Gavin Kerrigan, Padhraic Smyth, Mark Steyvers

NeurIPS 2021 · Label-efficient and robust learning › Recognition and long-tail · approach: Confidence calibration objectives · 7 citations (OpenAlex)

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 changedWhereMeasuredEffect
Combine calibrated model probabilities with human labelsResNet-164 · CIFAR-10Herror rate (%): 6.10 ± 0.38 -> 2.78 ± 0.15Improved
Combine calibrated model probabilities with human labelsVGG-19 · ImageNet-16H, noise 80error rate (%): 11.10 ± 0.60 -> 6.03 ± 0.54Improved
Add temperature scaling to the human-model combination (ablation)P+L on ResNet-164 · CIFAR-10Herror rate (%): 3.05 ± 0.23 -> 2.78 ± 0.15Improved
Use ensemble temperature scaling instead of temperature scaling (ablation)P+L on ResNet-164 · CIFAR-10Herror rate (%): 2.78 ± 0.15 -> 2.82 ± 0.23No clear effect
Use IMax binning instead of temperature scaling (ablation)P+L on ResNet-164 · CIFAR-10Herror rate (%): 2.78 ± 0.15 -> 2.85 ± 0.23No clear effect
Discard model probabilities and combine hard labels with confusion matrices (ablation)ResNet-164 plus human · CIFAR-10Herror rate (%): 2.78 ± 0.15 -> 4.71 ± 0.37Hurt
Fit the P+L combination with 10 rather than 5000 labeled datapoints (ablation)ResNet-164 P+L · CIFAR-10HECE (10^-2): 1.25 ± 0.18 -> 1.39 ± 0.18No clear effect
Add post-hoc recalibration to the combined predictions (ablation)ResNet-164 P+L · CIFAR-10HECE (10^-2): 1.39 ± 0.18 -> 0.84 ± 0.20Improved
Replace temperature scaling with IMax binning (ablation)P+L on VGG-19 · ImageNet-16H, noise 125error rate (%): 19.63 ± 0.70 -> 19.45 ± 0.62Improved
Use MAP rather than maximum-likelihood parameter estimation (ablation)Human-model combination · CIFAR-10H and ImageNet-16Hdata efficiencyImproved
Fit the combination without ground-truth labels using EM (ablation)P+L-EM · CIFAR-10H and ImageNet-16Herror rateMixed

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy