Deep neural networks

I am going through the hidden layer mission. I have got a question that so when the number of neurons is arbitrary, what exactly are they doing with the e.g. logit function? in particular, suppose I have 3 features, x_1, x_2, x_3 and have 1 hidden layer which has 6 neurons, what decision is each neuron making? and how are the weights of each decision calculated?
Thank you.