The steps involved in MIL are as follows:

1. Let's say we've a model parameterized by a parameter and we've a distribution over tasks . First we randomly initialize the model parameter .
2. Sample some batch of tasks from a distribution of tasks, that is, .
3. Inner loop: For each of the tasks in the sampled tasks, we sample a demonstration data—that is, . Now we compute loss and minimize the loss by performing gradient descent and we get the optimal parameters —that is,. Then, we also sample one more demonstration data for the meta training: .
4. Outer loop: Now we update our initial parameter using  by meta optimization, as follows:

5. Repeat steps 2 to 4 for n number of iterations.