In the last section, we saw how ADML works. We saw how we train our model with both clean and adversarial samples to obtain a better and robust model parameter θ that is generalizable across tasks. Now we will better understand ADML by coding them from scratch. For better understanding, we will consider a simple binary classification task. We randomly generate our input data and we train it with a single layer neural network and try to find the optimal parameter theta. Now, we will see step-by-step how exactly ADML works.

You can also check the code available as a Jupyter Notebook with an explanation here: https://github.com/sudharsan13296/Hands-On-Meta-Learning-With-Python/blob/master/06.%20MAML%20and%20it's%20Variants/6.7%20Building%20ADML%20From%20Scratch.ipynb.

First, we import all of the necessary libraries:

import tensorflow as tf
import numpy as np