Believing is Seeing


Abstract: Our world often starts with the premise that “Seeing is believing.” Why the very first time we conducted an experiment, we were told to look carefully at what the data told us to make meaningful inferences. Here, we argue that knowing a little bit about what we expect to see can go a long way in getting spectacular results; in terms of fidelity, sensitivity or dynamic range. I will use two examples, one where PIV data from a differentially heated rotating annulus experiment is used to constrain a very high dimensional fluid dynamical model in realtime, thus providing a basis for "observing the fluid using a model." I will then describe a second experiment in light field imaging, where knowing something about the prior statistics of the imaged flow enables for much more precise detection. I will finally turn to the Bayesian formulation inherent in these examples and my original statement, and illustrate few very quick and easy techniques applicable to imaging fluids without expensive inversions to accompany them.