Mode Classification in Fast-Rotating Stars using a Convolutional Neural Network: Model-Based Regular Patterns in $δ$ Scuti Stars

Oscillation modes in fast-rotating stars can be split into several subclasses, each with their own properties. To date, seismology of these stars cannot rely on regular pattern analysis and scaling relations. However, recently there has been the promising discovery of large separations observed in spectra of fast-rotating $δ$ Scuti stars: they were attributed to the island-mode subclass, and linked to the stellar mean density through a scaling law. In this work, we investigate the relevance of this scaling relation by computing models of fast-rotating stars and their oscillation spectra. In order to sort the thousands of oscillation modes thus obtained, we train a convolutional neural network isolating the island modes with 96\% accuracy. Arguing that the observed large separation is systematically smaller than the asymptotic one, we retrieve the observational $Δν- \overlineρ$ scaling law. This relation will be used to drive forward modelling efforts, and is a first step towards mode identification and inversions for fast-rotating stars.