Note 3 – Types of Learning

Learning with Different Output Space
1. Binary classification: $y = \{-1, +1\}$
2. Multiclass classification: $y = \{1, 2, ..., k\}$ .
3. Regression: $y = \mathbb{R}$ . (Or Bound Regression $y = \Theta$ ).
4. Structured learning: $y = \text{structures}$ .
Learning with Different Data Lable .
1. Supervised: all $y_n$ .
2. Unsupervised: no .
  1. clustering. (discrete).
  2. density estimation. (continuous).
  3. outlier detection.
3. Semi-supervised: part :
  1. expensive to label all the data
4. Reinforcement:
  1. learning with partial/implicit information.
Learning with Different Protocol
1. batch: all known data
2. online: sequential (passive) data
  1. Improve model while taking data
3. active: strategically-observed data
  1. Learn to ask a question. For example, ask when the confidence is low.
Learning with Different input Space
1. concrete: sophisticated and related physical meaning
2. raw: simple physical meaning
3. abstract: no or little physical meaning