Representation learning is a family of techniques in which models automatically learn useful features or internal representations from raw data. It reduces reliance on manually designed features and is central to modern deep learning.
See feature learning.