We will research new exciting methods, merging computer simulations,
Bayesian inference, deep learning and more generally machine learning,
to infer model parameters in stochastic models and quantify
uncertainties. Goal is to extract knowledge and perform predictions in challenging scenarios.
Mathematical and statistical models for real-world applications are complex. When constructing statistical and machine learning methodology to fit the model to data, we wish to obtain a realistic mathematical representation of the experiment under study. This means finding the "best" model parameters and quantify their uncertainties in a probabilistic way. However, what we typically face with realistically complex models is the inability to perform statistical inference exactly. This is typically due to the lack of a closed-form expression for the likelihood function of the model parameters. This affects both frequentist and Bayesian procedures. A change of paradigm is given by simulation-based "likelihood-free" approaches. In this case, the only requirement is the ability to simulate artificial data from am arbitrarily complex "computer simulator", which is a software implementation of the mathematical model. An important simulation-based approach is the class of Approximate Bayesian Computation (ABC) algorithms.
The goal of this project is to marry modern machine learning methodology such as deep neuronal networks, with simulation-based inference methods. This implies constructing new methodology for both areas, with the goal of creating novel "plug-and-play" engines for parameter inference and uncertainty quantification.
Special interest is geared towards learning the parameters of stochastic dynamical models (though not exclusively) such as stochastic differential equations and hidden Markov models (state space models). Moreover, while the project is of methodological nature, applications to real data case studies are foreseen, such as protein folding data, epidemic data and smart-city data such as such as precipitation, pollution data, temperature and humidity.
Here is a slightly more extended description with references.
PI is Umberto Picchini and PhD student Petar Jovanovski has been specifically hired for the project, which is an international collaboration withJes Frellsen (DTU Copenhagen), Andrew Golightly (Newcastle University, UK), Pierre-Alexandre Mattei (INRIA), and Samuel Wiqvist (Lund, Sweden).
The project is funded by the Chalmers AI Research centre and by the Swedish Research Council (Vetenskapsrådet).
Below is a link to the official University page of the PI Umberto Picchini. However a more informative page is his personal one.