Parameter averaging happens as follows:
- The master first initializes the neural network parameters based on the model configuration
- Then, it distributes a copy of the current parameters to each worker
- The training starts on each worker using its own subset of data
- The master sets the global parameters to the average parameters for each worker
- In those cases where there is more data to process, the flow repeats from Step 2
The following diagram shows a representation from Step 2 to Step 4:

Figure 10.3: Parameter averaging
In this diagram, W represents the parameters (weights and biases) in the network. In DL4J, this implementation ...