Ensemble Bayesian Neural Networks for Improved Out-of-Distribution Detection on Toy Datasets
Main Article Content
Abstract
Out-of-Distribution (OOD) detection is a vital problem in machine learning to identify inputs that differ significantly from training data. We propose a novel method combining ensemble Bayesian neural networks and a new OOD detection metric that integrates ensem- ble predictive entropy with the variance of uncertainty estimates. Evaluated on a synthetic toy dataset of 2D Gaussian clusters, the method demonstrates improved capability to dis- tinguish OOD samples by capturing complementary uncertainty information. We provide complete experimental code and visualizations.
Article Details
This work is licensed under a Creative Commons Attribution 4.0 International License.
This work is licensed under a Creative Commons Attribution 4.0 International License (CC BY 4.0). To view a copy of this license, visit https://creativecommons.org/licenses/by/4.0/
References
Yarin Gal and Zoubin Ghahramani. Dropout as a bayesian approximation: Representing model uncertainty in deep learning. In international conference on machine learning, pages 1050–1059. PMLR, 2016.
Dan Hendrycks and Kevin Gimpel. A baseline for detecting misclassified and out-of- distribution examples in neural networks. arXiv preprint arXiv:1610.02136, 2016.
Balaji Lakshminarayanan, Alexander Pritzel, and Charles Blundell. Simple and scalable predictive uncertainty estimation using deep ensembles. Advances in neural information processing systems, 30, 2017.
Shiyu Liang, Yixuan Li, and Rayadurgam Srikant. Enhancing the reliability of out-of- distribution image detection in neural networks. arXiv preprint arXiv:1706.02690, 2017.
Radford M Neal. Bayesian learning for neural networks, volume 118. Springer Science & Business Media, 2012.
Wendy S Parker. Ensemble modeling, uncertainty and robust predictions. Wiley interdisci- plinary reviews: Climate change, 4(3):213–223, 2013.
Lewis Smith and Yarin Gal. Understanding measures of uncertainty for adversarial example detection. arXiv preprint arXiv:1803.08533, 2018.
Andrew G Wilson and Pavel Izmailov. Bayesian deep learning and a probabilistic perspective of generalization. Advances in neural information processing systems, 33:4697–4708, 2020.