Representation Learning: A Fundamental Shift in Machine Learning

Representation learning is a transformative paradigm in machine learning, which is a ground breaking transformation from traditional feature engineering approaches to automatic discovery of features. In this comprehensive review, we look at the evolution of representation learning methods as applied to modern artificial intelligence systems, as well as their implications for both practitioners and researchers. With popular machine learning applications getting more complex, a need to automatically learn high level representations out of raw data is important for making the field go beyond manual feature engineering.