Logistic Regression with a Neural Network mindset We are going to build a logistic regression from a neural network point of view to separate 2 classes from the MNIST dataset. The neural network will have 28x28 = 784 input neurons, no hidden layers and 1 output with a sigmoid as activation function. The output (𝑦) will take values from 0 to 1 and can be thresholded at 0.5 to predict the classes.