Speaker
Will Cunningham
Description
Is it worthwhile to study quantum gravity using deep
learning? In causal set quantum gravity, it is useful for
measuring discrete observables in Monte Carlo simulations.
We have found that though there does not yet exist a general
analytical expression for a causal set's manifold dimension,
we can train a deep neural network in under five minutes to
identify the correct value with over 99% accuracy.
Remarkably, when trained with only 100-element Minkowski
causal sets, the network can still correctly identify the
dimension for those generated from other manifolds and those
with larger sizes. We will discuss the types of problems for
which deep learning is well suited, and also look at how one
constructs a supervised learning algorithm using the
TensorFlow package for Python. Finally, we will conclude by
discussing how these methods may be trivially extended to
study other discrete geometric properties.