Victor Elvira awarded Leverhulme Research Fellowship entitled "Automatic and self-assessed Bayesian inference for real-world dynamic systems"
Victor Elvira has been awarded a Leverhulme Research Fellowship entitled "Automatic and self-assessed Bayesian inference for real-world dynamic systems" worth almost £60,000. In relevant real-word applications complex systems are described through dynamic models that relate observed data to a hidden state that must be estimated. In the Bayesian approach, the posterior distributions of those states are approximated with sequential Monte Carlo (SMC). Unfortunately, SMC methods introduce approximation errors, and the models never describe the system perfectly, resulting in poor inferential performance. This project will focus on two methodological developments: (a) a novel framework for automatic self-assessment of the model and the inference, detecting the malfunctioning, identifying the cause, and triggering immediate responses; and (b) new computational methods for allocating the computational resources efficiently.