Differentiable Systematic Resampling for Variational Sequential Monte Carlo

Particle filters are a standard tool for nonlinear state estimation, but their resampling step is discrete, preventing gradient-based learning in variational sequential Monte Carlo.