Learning with different output space
1.classification
2.regression
3.structured:such as judging whether a sentence is legal
Learning with different reinforcement
1.supervised:every 
2.un/semi-supervised: no/some labeled data
3.reinforcement:learn with ‘partial/implicit information’(often sequentially)
Learning with different protocol
1.batch:input, ouput are confirmed and sent to machine a large amount
2.online:hypothesis ‘improves’ through receiving data instances sequentially
3.active:improve hypothesis with fewer labels(hopefully) by asking questions strategically
