Terms

1. Multi

https://www.coursera.org/lecture/machine-learning-projects/multi-task-learning-l9zia

multi-class	multi-label	multi-task
one sample has one true label	one sample has a set of labels	generalize multi-label
softmax	log-loss
ex: MNIST	ex: document has multiple topics: sport, military, and/or finance.
ex: a fruit should be apple, orange or pearl

Multi-label can be replaced by several binary classifiers. However, multi-label shares features and reduces training resource.

Multi-label is better for balanced dataset.