Sequential design of point-wise measurements based on a deep generative model

We present a new method to perform 2D measurements based on planning pointwise measurements. This can replace the raster scan for many scientific experiments. The proposed method is based on a deep generative model to generate multiple possible reconstructions given an arbitrary number of scattered measurements, and information theory is used to choose the optimal next location for the point-wise measurement. The proposed method is not domain-specific and could be used for any type of 2D measurement as long as training data is available.