Qu&Co comments on this publication:

In the 2016 US presidential elections, many of the professional polling groups had overestimated the probability of a Clinton victory. Multiple post-election analyses concluded that a leading cause of error in their forecast models was a lack of correlation between predictions for individual states. Uncorrelated models, though much simpler to build and train, cannot capture the more complex behavior of a fully-connected system. Accurate, reliable sampling from fully-connected graphs with arbitrary correlations quickly becomes classically intractable as the graph size grows. In this paper, Henderson et al. show an initial implementation of quantum-trained Boltzmann machine used for sampling from correlated systems. They show that such a quantum-trained machine is able to generate election forecasts with similar structural properties and outcomes as a best in class modeling group.