What is Probabilistic AI?
Probabilistic AI involves embedding probabilistic models, probabilistic reasoning and measures of uncertainty within AI methods. The current AI toolbox is already heavily based on probabilistic models, such as softmax classifiers, transformers for generative AI, and diffusion models. Many algorithms used within AI are probabilistic in nature. Also, probabilistic reasoning has been shown to give a theoretically optimal way to reason and track uncertainty in AI models, and gives a natural framework to incorporate uncertainty in decisions.
What is the Prob_AI Hub?
The Hub aims to bring together researchers in Applied Mathematics, Probability and Statistics to work on challenges in probabilistic AI. This is motivated by the observation that many fundamental problems in AI would benefit from input from these disciplines: be it in understanding deep neural networks by looking at their (dynamical systems and probabilistic) asymptotic behaviour; or improving and generalising diffusion-based generative AI with input from numerical and stochastic analysis; to making AI models more interpretable and reliable by incorporating known constraints or physical properties.