Efficient Evaluation-Time Uncertainty Estimation by Improved Distillation
This work addresses the need for efficient uncertainty estimation in machine learning, which is crucial for reliable deployment in real-world applications, but it appears incremental as it builds upon existing distillation methods.
The paper tackles the problem of obtaining computationally-efficient uncertainty estimates with deep networks by proposing a modified knowledge distillation procedure, achieving state-of-the-art uncertainty estimates for both in and out-of-distribution samples.
In this work we aim to obtain computationally-efficient uncertainty estimates with deep networks. For this, we propose a modified knowledge distillation procedure that achieves state-of-the-art uncertainty estimates both for in and out-of-distribution samples. Our contributions include a) demonstrating and adapting to distillation's regularization effect b) proposing a novel target teacher distribution c) a simple augmentation procedure to improve out-of-distribution uncertainty estimates d) shedding light on the distillation procedure through comprehensive set of experiments.