loss function

• loss over entire dataset = avg loss over examples $L=\frac{1}{N}{\Sigma }_{i=1}^N{L}_i(y_i,{\hat{y}}_i)$

examples

• L2 loss: squared error $(y_i-\widehat{y_i}{)}^2$
  • not robust to outliers
• L1 loss $|y_i-{\hat{y}}_i|$

to minimize loss → gradient descent