Learning Intelligent Machine (LIM) lab’s paper is accepted to 35th Conference on Neural Information Processing Systems (NeurIPS) 2021, one of the top-3 conferences for artificial intelligence and machine learning.
Bootstrapping has been a primary tool for ensemble and uncertainty quantification in machine learning and statistics.
However, due to its nature of multiple training and resampling, bootstrapping deep neural networks is computationally burdensome; hence it has difficulties in practical application to uncertainty estimation and related tasks. To overcome this computational bottleneck, we propose a novel approach called NeuBoots (Neural Bootstrapper), which learns to generate bootstrapped neural networks through single model training.
Our empirical results show that NeuBoots outperforms other uncertainty quantification methods under a much lower computational cost without losing the validity of bootstrapping.
“Neural Bootstrapper” by Minsuk Shin, Hyungjoo Cho, Hyun-seok Min, Sungbin Lim*
Note: *Corresponding Author