Biophysics

Active learning of deep ensembles to predict behavior of protein sequences

Can we efficiently train neural networks to learn protein properties with very few training examples? Can the machine guide which expensive data would be most useful to obtain?