Structural Risk Minimization
Minimize structural risksIt is the induction principle in machine learning and is often used as a strategy to prevent overfitting.
SRM Principles
Structural risk = empirical risk + confidence risk
In optimization theory, the minimum structured risk is mainly the empirical risk on the sample. Under the premise of preventing overfitting, the confidence risk can be minimized by adding regular terms.
When the sample size is large enough, the empirical risk approaches the structural risk. Since minimizing the empirical risk can ensure the learning effect, it is widely used in reality.
SRM Applications
In Bayesian estimation, maximum a posteriori probability estimation is structured risk minimization.
The model follows the conditional probability distribution, the loss function is the logarithmic loss function, the complexity of the model is represented by the model prior probability, and the structural risk minimization is the maximum a posteriori probability estimation.