MAML Inner Loop Step

Problem

Run the MAML inner loop on one task by applying a sequence of gradient steps.

Starting from params, apply theta = theta - lr * grads for each gradient vector in grad_steps and return the adapted parameters. An empty grad_steps returns a copy of params.

Hint

These are the fast adaptation steps MAML differentiates through when computing the meta-gradient.

Test cases

case 1

in: [[1,2],[[0.5,-0.5]],0.1]

expected: [0.95,2.05]

case 2

in: [[0],[[1],[1]],0.5]

expected: [-1]

case 3

in: [[1],[],1]

expected: [1]

case 4

in: [[],[],0.5]

expected: []

Starter codePython

def maml_inner_loop(params, grad_steps, lr):
    # Your code here
    pass

4 test cases · run your code against them in the editor