Uncertainty Estimation in AI

The Accelerate Programme for Scientific Discovery, based in the Department of Computer Science and Technology, offers support for researchers across the University to use AI in their research. We are pleased to announce that our AI training courses & workshops for this term are open for booking!

Join us for courses exploring key topics in AI as well as hands-on workshops to apply your AI skills. 

Uncertainty Estimation in AI

Neural networks have had success in science and engineering predictive tasks such as computer vision, biology, medicine, and weather. To safely use (make decisions based on) the predictions of neural networks, the predictions need to have reliable uncertainty estimates. Yet, despite their importance, reliable predictive uncertainty estimates are often overlooked in practice. This might be due to the fact that producing such estimates is difficult. By providing an honest overview of modern uncertainty estimation methods, the participants can decide whether to adopt them in their own research.

The aim of the workshop is two-fold. First, to give a gentle introduction to what uncertainty is, the how to estimate it, and how to evaluate it in the context of neural networks. Secondly, to have participants get hands-on experience by implementing and evaluating uncertainty methods, such as Deep Ensembles, Variational Inference, and Laplace approximations.

This workshop will run over two mornings (day 1: lectures and theory, day 2: hands on practice). The workshop is open to all postgraduate students and research staff.

Open to all postgraduate students and staff, our courses are designed to support you at all stages of implementing AI in your research. 

Registration is required to attend. Refreshments will be provided.