Deep Neural Network Calibration by Reducing Classifier Shift with Stochastic Masking
This addresses the issue of unreliable confidence estimates in safety-critical scenarios like autonomous driving and healthcare, though it appears incremental as it builds on existing classifier modification approaches.
The paper tackled the problem of poor calibration in deep neural networks, which leads to unreliable confidence estimates in safety-critical applications, by proposing MaC-Cal, a mask-based classifier calibration method that uses stochastic sparsity to improve alignment between confidence and accuracy, achieving superior calibration performance and robustness under data corruption.
In recent years, deep neural networks (DNNs) have shown competitive results in many fields. Despite this success, they often suffer from poor calibration, especially in safety-critical scenarios such as autonomous driving and healthcare, where unreliable confidence estimates can lead to serious consequences. Recent studies have focused on improving calibration by modifying the classifier, yet such efforts remain limited. Moreover, most existing approaches overlook calibration errors caused by underconfidence, which can be equally detrimental. To address these challenges, we propose MaC-Cal, a novel mask-based classifier calibration method that leverages stochastic sparsity to enhance the alignment between confidence and accuracy. MaC-Cal adopts a two-stage training scheme with adaptive sparsity, dynamically adjusting mask retention rates based on the deviation between confidence and accuracy. Extensive experiments show that MaC-Cal achieves superior calibration performance and robustness under data corruption, offering a practical and effective solution for reliable confidence estimation in DNNs.