Skip to main content
Article
Nonparametric Uncertainty Quantification for Single Deterministic Neural Network
Advances in Neural Information Processing Systems
  • Nikita Kotelevskii, Skolkovo Institute of Science and Technology
  • Aleksandr Artemenkov, Skolkovo Institute of Science and Technology
  • Kirill Fedyanin, Technology Innovation Institute
  • Fedor Noskov, Skolkovo Institute of Science and Technology
  • Alexander Fishkov, Skolkovo Institute of Science and Technology
  • Artem Shelmanov, AIRI & Mohamed bin Zayed University of Artificial Intelligence
  • Artem Vazhentsev, Skolkovo Institute of Science and Technology
  • Aleksandr Petiushko, Lomonosov Moscow State University
  • Maxim Panov, Technology Innovation Institute
Document Type
Conference Proceeding
Abstract

This paper proposes a fast and scalable method for uncertainty quantification of machine learning models' predictions. First, we show the principled way to measure the uncertainty of predictions for a classifier based on Nadaraya-Watson's nonparametric estimate of the conditional label distribution. Importantly, the proposed approach allows to disentangle explicitly aleatoric and epistemic uncertainties. The resulting method works directly in the feature space. However, one can apply it to any neural network by considering an embedding of the data induced by the network. We demonstrate the strong performance of the method in uncertainty estimation tasks on text classification problems and a variety of real-world image datasets, such as MNIST, SVHN, CIFAR-100 and several versions of ImageNet.

Publication Date
12-1-2022
Comments

IR conditions: non-described

Access available on NeurIPS Proceedings site

Citation Information
N. Kotelevskii, et al, "Nonparametric Uncertainty Quantification for Single Deterministic Neural Network", in Advances in Neural Information Processing Systems,vol. 35, Dec 2022. https://proceedings.neurips.cc/paper_files/paper/2022/file/eb7389b039655fc5c53b11d4a6fa11bc-Paper-Conference.pdf