Nonlinear dynamics reconstruction with neural networks of chaotic time-delay communication systems

The nonlinear dynamics of semiconductors lasers with electro‐optical feedback is reconstructed using a new type of modular neural network. It is shown that these neural networks are more efficient than the standard feedforward ones. We find that the difficulty to reconstruct the nonlinear dynamics increases with the feedback strength and with the number of the delays, but not with the delay time. Finally, the neural networks are used to extract the messages encoded in a chaos‐based communication system.