Uncertainty in neural networks
How to be certain about uncertainty in machine learning? In my talk, I’d like to cover different types of uncertainties with short practical examples of how to detect and tackle them. We will go through the aleatoric error which will show you how to detect hard cases in your data, epistemic error – thanks to which you can tell if a model knows example which you provided to such degree that you can trust its prediction and distributional error – which is signaling if the example is completely unknown. Thanks to this knowledge one can easily augment her / his models in order to provide more robust solutions.