Revisiting the Calibration of Modern Neural Networks
This addresses the problem of predictive uncertainty estimation for safe AI applications, providing insights into how architecture affects calibration, though it is incremental as it builds on prior observations.
The study revisited the calibration of modern neural networks, finding that recent state-of-the-art image classification models, especially non-convolutional ones, are among the best calibrated, with trends like decay under distribution shift being less pronounced.
Accurate estimation of predictive uncertainty (model calibration) is essential for the safe application of neural networks. Many instances of miscalibration in modern neural networks have been reported, suggesting a trend that newer, more accurate models produce poorly calibrated predictions. Here, we revisit this question for recent state-of-the-art image classification models. We systematically relate model calibration and accuracy, and find that the most recent models, notably those not using convolutions, are among the best calibrated. Trends observed in prior model generations, such as decay of calibration with distribution shift or model size, are less pronounced in recent architectures. We also show that model size and amount of pretraining do not fully explain these differences, suggesting that architecture is a major determinant of calibration properties.