The variance of algorithms can be defined as follows:

In the previous equation, implies the geometric mean of .

We can use any of these inequality measures to calculate the task bias. So, once we calculate the task bias using this inequality measure, we can minimize the bias by plugging our inequality measure into the meta objective. So, we can rewrite our meta objective as follows:

In the previous equation, represents our inequality measure and is the balancing coefficient.