End-to-End Learning of Gaussian Mixture Proposals Using Differentiable Particle Filters and Neural Networks Conference Paper