Figure 2
From: Model selection to achieve reproducible associations between resting state EEG features and autism
ML modeling pipeline diagram. First, an autoencoder is trained to approximate the data \(\:\varvec{x}\) from a low-dimensional latent space using the Itakura-Saito divergence (IS divergence) as the reconstruction objective. After training the autoencoder, both the data \(\:\varvec{x}\) and data reconstruction \(\:\widehat{\varvec{x}}\) are used in a deterministic transform based on signal processing relationships to produce a non-negative, transformed version of the data \(\:\stackrel{\sim}{\varvec{x}}\). A logistic regression classifier is then trained to predict either diagnosis of autism or neurotypical (NT) using the transformed version of the data \(\:\stackrel{\sim}{\varvec{x}}\) as input. Finally, after logistic regression classifier training the logistic regression coefficients \(\:\varvec{\omega\:}\) are taken and used to interpret the learned associations with autism diagnosis or NT diagnosis.
