Speaker
Description
We apply the Transformer architecture to explore the vast theory space of two-dimensional Rational Conformal Field Theories (RCFTs) spanned by their tensor products. Our data features the primary conformal dimension and the central charge of the RCFTs from different families, most of which come from the affine Kac-Moody algebra. Constructing the tensor product models based on RCFTs is often controllable due to their exact-solvability but the reverse-engineering task is not. We thus tackle the following inverse problem: given limited information about the spectra of a theory, can we decompose the theory in terms of RCFTs. The problem can be formulated in a standard sequence-to-sequence task suitable for the Transformer model. We design several experiments that work on different families of RCFTs and their tensor products at arbitrary values of central charge. In most cases, we obtain a high accuracy (>90%) using Transformers with small parameters. Our work shows that the Transformer model is able to work on RCFT data and potentially extends our knowledge about the theory space of two-dimensional RCFTs that can be infinitely large.
