Supervised learning of few dirty bosons with variable particle number

We investigate the supervised machine learning of few interacting bosons in optical speckle disorder via artificial neural networks. The curve shows an approximately universal power-law scaling for different particle numbers and interaction strengths. introduce a network architecture that can be trained tested on heterogeneous datasets including numbers. This provides accurate predictions all system sizes included training set and, by design, is suitable to attempt extrapolations (computationally challenging) larger sizes. Notably, novel transfer-learning strategy implemented, whereby systems substantially accelerated made consistently many small-size instances.