In this work, we perform a wide variety of experiments with different deep
learning architectures on datasets of limited size. According to our study, we
show that model complexity is a critical factor when only a few samples per
class are available. Differently from the literature, we