What is the meaning of 3-layer NN, 300+100 hidden units?

Question

I am new to NN. I was getting my hands on MNIST Handwritten Dataset. They have specified the error percentage for every method they adopted in a tabular form. Here is link to the page. In NN section they have specified like :

3-layer NN, 300+100 hidden units

2-layer NN, 300 hidden units, mean square error

what does 3 layers mean ? is it

InputLayer+HiddenLayer+OutputLayer

OR

InputLayer + 3 Hidden Layer+ outputLayer

Also what is the meaning of 300+100 hidden units. Specifing two numbers 300+100 indicates one hidden layer should contain 300 units and next HiddenLayer should contain 100 units .

If this what is meant then why 3 Layer NN ?

Answer 1

The input layer isn't counted in this notation, so 3 layer mean 2 hidden layers and 1 output layer. So in this case :

3-layer NN, 300+100 hidden units

means the following architecture :

Input layer
Hidden layer of 300 units
Hidden layer of 100 units
Output layer