Accuracy-Preserving Calibration via Statistical Modeling on Probability Simplex
This addresses the need for reliable uncertainty estimates in classification models, particularly for high-accuracy pre-trained networks, though it is incremental as it builds on existing probabilistic calibration approaches.
The paper tackles the problem of calibrating deep neural network classifiers without sacrificing their accuracy, proposing a method that uses the Concrete distribution on the probability simplex and theoretically proving its optimality, resulting in outperformance over previous methods in benchmarks.
Classification models based on deep neural networks (DNNs) must be calibrated to measure the reliability of predictions. Some recent calibration methods have employed a probabilistic model on the probability simplex. However, these calibration methods cannot preserve the accuracy of pre-trained models, even those with a high classification accuracy. We propose an accuracy-preserving calibration method using the Concrete distribution as the probabilistic model on the probability simplex. We theoretically prove that a DNN model trained on cross-entropy loss has optimality as the parameter of the Concrete distribution. We also propose an efficient method that synthetically generates samples for training probabilistic models on the probability simplex. We demonstrate that the proposed method can outperform previous methods in accuracy-preserving calibration tasks using benchmarks. The code is available at https://github.com/ToyotaCRDL/SimplexTS.