Empirical Risk Minimisation Algorithm

machine-learning

Definition

Empirical Risk Minimisation Algorithm

A learning algorithm is called empirical risk minimising if it always returns a hypothesis $h_S$:

$$h_S\in \arg \underset{h\in \mathcal{H}}{\min }{R}_S(h)$$

where:

• $S\sim D$ is the training sample sampled from an unknown independently and identically distributed distribution $D$.
• $R$ is the empirical risk.
• $h_S$ is the hypothesis that results in the lowest possible empirical risk.