Structural Risk
Structural risksIt is a compromise between empirical risk and expected risk. Usually, structural risk is obtained by adding a regularization term after the empirical risk function.
Concept Explanation
- Confidence risk: the error of an untrained classifier when classifying an unknown sample;
- Empirical risk: The error obtained by a trained classifier after fully classifying the training samples;
- Structural risk: confidence risk + experience risk.
The significance of structural risk
Structural risk minimization is an extension of empirical risk minimization. The smaller the empirical risk, the more complex the model decision function is and the more parameters it contains. When the empirical risk function is small to a certain extent, overfitting will occur.
In order to ensure overfitting and minimize the regularization term, two minimization functions are required.
The structural risk function integrates the two and ensures that the complexity of the empirical risk function and the model decision function is minimized. The structural risk function is then minimized to achieve the optimization goal.