I was reading a paper Multi-Agent Reinforcement Learning for Adaptive User Association in Dynamic mmWave Networks and I was stuck understanding the deep neural network architecture that was used. The authors gave it in Fig. 3 (on top of page 6) and they state the following (on page 9):
This architecture comprises 2 multi-layers perceptron (MLP) of 32 hidden units, one RNN layer (a long short memory term - LSTM) layer with 64 memory cells followed by another 2 MLPs of 32 hidden units. The network then branches off in two MLPs of 16 hidden units to construct the duelling network.
According to Fig. 3 there is one MLP, one RNN and one MLP. So why the authors said 2 MLPs?
Assuming it is 2 MLPs, does this mean we have 2 hidden layers of 32 neurons each? So, at the end we will have:
one input layer - one hidden layer with 32 neurons - another hidden layer with 32 neurons - one RNN layer with 64 cells - one hidden layer with 32 neurons - another hidden layer with 32 neurons - one hidden layer with 16 neurons - another hidden layer with 16 neurons - one output layer.