Towards a Cross-Research Platform for Hosting Bayesian Data-fitting Tools
This workshop will bring together stakeholders from research and industry to facilitate knowledge exchange in data-fitting problems and Bayesian samplers.
Scientists and engineers enhance their understanding of the world by fitting models to data. The model parameters are however inherently uncertain due to observational errors in the data, and structural uncertainties in the model. Bayesian sampling methods offer an approach to quantifying the uncertainty in model parameters by inferring the full posterior probability distribution (prior probability multiplied by the likelihood) of the model as a function of its parameters. In static systems, the models are usually quick to run for each set of parameters. Real systems such as interacting galaxies, the weather, or flu epidemics often change rapidly with time, requiring more complex models. Applying Bayesian methods to these problems can be computationally prohibitive.
This free virtual one-day workshop will bring together 30+ stakeholders in both research and industry to facilitate knowledge exchange in the quantification of uncertainty in complex data-fitting problems. There will be a combination of pedagogical talks, research talks, and discussion sessions that will cover everything from identifying complex data-fitting problems for which uncertainties need to be quantified to developing state-of-the-art Bayesian sampling methods. This will pave the first steps towards creating a cross-research platform for testing a wide range of samplers on various complex data-fitting problems.
The speakers contributed either full short papers or abstracts to a conference proceedings which is available here.
