Neural Clamping: Joint Input Perturbation and Temperature Scaling for Neural Network Calibration
This addresses the problem of model calibration for users needing reliable confidence estimates in deep learning applications, representing an incremental improvement over existing post-processing techniques.
The paper tackles neural network calibration by proposing Neural Clamping, a post-processing method that uses joint input perturbation and temperature scaling, and shows it significantly outperforms state-of-the-art methods on datasets like BloodMNIST, CIFAR-100, and ImageNet.
Neural network calibration is an essential task in deep learning to ensure consistency between the confidence of model prediction and the true correctness likelihood. In this paper, we propose a new post-processing calibration method called Neural Clamping, which employs a simple joint input-output transformation on a pre-trained classifier via a learnable universal input perturbation and an output temperature scaling parameter. Moreover, we provide theoretical explanations on why Neural Clamping is provably better than temperature scaling. Evaluated on BloodMNIST, CIFAR-100, and ImageNet image recognition datasets and a variety of deep neural network models, our empirical results show that Neural Clamping significantly outperforms state-of-the-art post-processing calibration methods. The code is available at github.com/yungchentang/NCToolkit, and the demo is available at huggingface.co/spaces/TrustSafeAI/NCTV.