Isotonic Regression
A non-parametric probability calibration method that adjusts model confidence scores by fitting a monotonic regression function. It requires an independent validation set to estimate parameters.
Core Facts
Timeline (last 90 days)
Common post-calibration techniques include Temperature Scaling, Platt Scaling, and Isotonic Regression, which remap raw outputs to more reliable probability estimates by fitting correction functions on an independent validation set without retraining the model.
Technical methods for improving calibration include post-processing approaches such as temperature scaling and isotonic calibration.
All Facts (3)
Neural networks have an overconfidence defect: even when inputs fall outside the training distribution, the Softmax layer may still output inflated confidence scores.
75%UnverifiedCommon post-calibration techniques include Temperature Scaling, Platt Scaling, and Isotonic Regression, which remap raw outputs to more reliable probability estimates by fitting correction functions on an independent validation set without retraining the model.
70%UnverifiedTechnical methods for improving calibration include post-processing approaches such as temperature scaling and isotonic calibration.
50%