Schematic representation of the neural network architecture used for classifying CEERS galaxies in this work. A first CNN (G_f(θ_f)) is fed with both labeled and unlabeled CANDELS and CEERS stamps respectively. The computed features are then used as input for two additional CNNs: a discriminator (G_d(θ_d)), which learns to distinguish stamps coming from the two data sets, and a classifier (G_y(θ_y)) which provides a classification in four main morphological classes. More details about the training strategy can be found in the text.