 Uniform stability of a learning algorithm

A learning algorithm has uniform stability $\beta$ for a loss when replacing one training example changes the loss at every test example by at most $\beta$.