ERM — Empirical Risk Minimization

Fact

Learning by making average training mistakes as small as possible.

In Plain Words

ERM is like a kid studying only the practice test. They get those answers perfect, then hope the real test is not mean.

It is a basic goal in Supervised learning. The model first fits the training examples. Then we worry about new ones.

Related Concepts

• Supervised Learning
• Stochastic Gradient Descent
• Bias-Variance Tradeoff
• Regularization

Supervised Learning
ERM is one common goal in Supervised learning.

SGD
SGD often lowers empirical risk one small step at a time.

Bias-Variance Tradeoff
Pushing empirical risk too low can hurt new answers.

Regularization
Regularization holds ERM back, so the model does not just memorize training examples.