We are pleased to share that a research paper co-authored by Dr. Maxim Panov, senior research scientist, and Kirill Fedyanin, research engineer at the AI and Digital Science Research Center (AIDRC), titled “Nonparametric Uncertainty Quantification for Single Deterministic Neural Network”, has been accepted for publishing at NeurIPS 2022, the 36th Conference on Neural Information Processing Systems, that runs from November 28 – December 9 in New Orleans, US.
Founded in 1987, the NeurIPS conference is a premier international AI event that promotes knowledge exchange of research advances in AI and Machine Learning (ML). The annual interdisciplinary conference complies with the highest ethical standards to ensure access for a diverse and inclusive community. The conference encompasses invited talks, demonstrations, symposia, and oral and poster presentations of refereed papers.
The AIDRC team’s paper explored a fast and scalable method for uncertainty quantification of machine learning models' predictions. It proposed a theoretically grounded nonparametric approach that allows the explicit disentanglement of aleatoric and epistemic uncertainties. As such, the resulting method can be applied to any neural network. The results also demonstrated 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.
The efforts of Dr. Panov and Fedyanin as well as their co-authors, inevitably augment the next generation of AI and ML, establishing a platform for the global AI fraternity to share knowledge and exchange academic discourse. The acceptance also marks a recognition of the caliber of the UAE’s scientific community in formulating cutting-edge research for the advancement of AI and ML in the country and the wider region.
Congratulations once again to Dr. Maxim Panov and Kirill Fedyanin! We look forward to more success and industry recognition in the future!
