ML: Loss functions

Notes.

Classification:

1. Cross-entropy loss: -sum(plogpi)

2. Hinge loss: SVM, max(0, 1-yf(x))

3. Exponential loss: Adaboost, exp(-betayf(x))

4. Triplet loss: google 2015 Facenet

Regression:

1. L1 loss/MAE (mean absolute loss): 1/nsum(abs(y-yhat))

2. L2 loss/MSE (mean squared loss): 1/nsum(y-yhat)**2