ITP |
|
Teilchentee SS 2021
Date: 10.06.2021
Speaker: James Halverson
Phenomenology of Strings and Neural Networks
One might describe particle or condensed matter phenomenology to a statistician as the careful design of a non-Gaussian statistical model over functions (path integral) to account for observed interactions. String theory gives many such statistical models, i.e. low energy effective theories, and I'll review what sorts of surprises arise relative to standard expectations of a model builder, under the assumption that we are drawing from a uniform distribution on known string EFTs; ALPs and dark gauge sectors abound, leading to cosmological problems and opportunities. The map from string theoretical data to string EFT is cumbersome, though, so I'll describe how one might simulate it via a learned random tensor approximation with a deep convolutional Wasserstein GAN. Finally, having demonstrated ways in which strings broaden our traditional notions of phenomenology, I'll go further by introducing a phenomenology of neural network architectures.
