e7d019329e662fe4685be505befca3bb-Paper-Conference.pdf

Inductive biases encoding known data symmetries are key to make deep learning models generalize in high-dimensional settings such as computer vision, speech processing and computational neuroscience, just to name a few.