This project will leverage artificial neural networks to automatically build various components of particle filters.

点击翻译按钮获取中文摘要

Particle filters are a family of flexible algorithms using Bayesian Monte Carlo methods to sequentially update the probability distribution of internal states of a dynamic system from noisy measurements. The methods are statistically consistent and are widely used in applications such as computer vision, target tracking, finance, navigation, and robotics. A particular challenge for particle filters is the need to specify the dynamic and measurement models that simulate state dynamics and their relation to measurements. This becomes non-trivial for practitioners when dealing with complex environments. This project will leverage artificial neural networks to automatically build various components of particle filters. Replacing heuristic models in particle filters with data-driven ones would make them an extremely powerful tool in applications central in the digitisation and the digital NMI agenda such as autonomous driving under different road conditions. The proposed differentiable particle filter (DPF) framework lies on the intersection of dynamic systems and robust AI, attempting toanswer the fundamental question of how we analyse dynamic systems based on observed data, rather than through an analytic, heuristic approach. It links not only closely to current research activities on Bayesian inference for complex systems, but also NMS activities such as cross theme CAV project. Project outcomes are expected to be applicable to a wide range of applications in the digital sector