We'll always be biased until we directly observe Dark Matter ;) There's another gotcha: if we are inferring the hidden from observable properties, might not th

We'll always be biased until we directly observe Dark Matter ;) There's another gotcha: if we are inferring the hidden from observable properties, might not the charge of "feature engineering" also creep into the model? Why human observable properties of the galaxies and not nature's implicit representations? Pretty eye-opening to think with enough supervised training data humanity approaches a GAN for cosmology! Discovering Symbolic Models from Deep Learning with Inductive Biases https://deepmind.com/research/publications/2020/Discovering-Symbolic-Models-from-Deep-Learning-with-Inductive-Biases https://deepmind.com/research/publications/2020/Discovering-...