13/03/2026 - 14:30 - ISBA C115 -
Will give a presentation on :
Abstract:
Advancements in artificial intelligence (AI) and deep learning have led to neural networks being used to generate lightning-speed answers to complex questions, to paint like Monet, or to write like Proust. Leveraging their computational speed and flexibility, neural networks are also being used to facilitate fast, likelihood-free statistical inference. However, it is not straightforward to use neural networks with data that for various reasons are incomplete, which precludes their use in many applications. In the first part of the talk, I will describe a novel neural Bayes inference approach that is based on the Monte Carlo expectation-maximization (EM) algorithm. The proposed neural EM approach is likelihood-free, substantially faster than the conventional EM algorithm when conditional simulation from the model is feasible (as it does not require numerical optimization at each iteration), and more statistically efficient than a simpler alternative “masking approach”. The proposed inference approach will be illustrated using simulated data and an application to Arctic sea-ice data, analyzed using a hidden Potts model with an intractable likelihood function. If time allows, in the second part of the talk, I will then explain how score-based diffusion models can be used to perform fast and accurate conditional simulation from spatial models when this is theoretically infeasible or computationally expensive, such as with spatial extremes models.