Data-driven deep density estimation

Density estimation plays a crucial role in many data analysis tasks, as it infers continuous probability density function (PDF) from discrete samples. Thus, is used tasks diverse analyzing population data, spatial locations 2D sensor readings, or reconstructing scenes 3D scans. In this paper, we introduce learned, data-driven deep (DDE) to infer PDFs an accurate and efficient manner, while being independent of domain dimensionality sample size. Furthermore, do not require access the original PDF during estimation, neither parametric form, nor priors, form This enabled by training unstructured convolutional neural network on infinite stream synthetic PDFs, unbound amounts generalize better across deck natural than any finite will do. hope that our publicly available DDE method be beneficial areas analysis, where models are estimated observations.