Uncertainty Quantification: Scaling up Bayesian Neural Networks with Neural Networks

Date:

  • Proposed a fast training Bayesian Neural Network (FBNN) with an efficient sample collection process by a Neural Network emulator, while maintaining the quality in terms of prediction accuracy and uncertainty quantification
  • Accelerated the training process using a collection-emulation-sampling strategy to enhance computational efficiency
  • Achieved 95 times increment in the minimal effective sample size compared to the existing benchmark methods