Uncertainty in deep learning, Inbar Naor, Taboola

Understanding what a model doesn’t know is important both from a practitioner point of view and for the end users of many different machine learning applications. However, traditional machine learning methods that handle uncertainty, such as Gaussian process, do not scale well, while deep learning models are typically used to get point estimates, leaving this kind of information out. In this talk I will talk about uncertainty, why it’s important and how can we incorporate bayesian thinking into deep learnin