Scalable Representation Learning: On Data-Scarcity, Uncertainty and Symmetry