Hands-On Meta Learning with Python

Single layer neural network

We use a neural network with only a single layer for predicting the output:

a = np.matmul(X, theta)YHat = sigmoid(a)

So, we use Meta-SGD for finding this optimal parameter value theta, and learning rate and gradient update direction that's generalizable across tasks. So, for a new task, we can learn from a few data points in less time by taking fewer gradient steps.