Machine learning in the string landscape

Jonathan Carifio, James Halverson, Dmitri Krioukov, Brent D. Nelson
Journal of High Energy Physics
2017: 157
September 28, 2017

Abstract

We utilize machine  learning to study the string landscape. Deep data dives and conjecture  generation are proposed as useful frameworks for utilizing machine learning  in the landscape, and examples of each are presented. A decision tree  accurately predicts the number of weak Fano toric threefolds arising from  reflexive polytopes, each of which determines a smooth F-theory  compactification, and linear regression generates a previously proven  conjecture for the gauge group rank in an ensemble of 43×2.96×10755  F-theory compactifications. Logistic regression generates a new conjecture  for when E6 arises in the large ensemble of F-theory compactifications, which  is then rigorously proven. This result may be relevant for the appearance of  visible sectors in the ensemble. Through conjecture generation, machine  learning is useful not only for numerics, but also for rigorous results.

Related publications