Adapted Deep Embeddings: A Synthesis of Methods for $k$-Shot Inductive Transfer Learning