A novel architecture and learning algorithm for a multilayered echo state network (ML-ESN). The addition of multiple layers of reservoir are shown to provide a more robust alternative to conventional RC networks.

Malik, Z. K., Hussain, A., & Wu, Q. J. (2017). Multilayered echo state machine: 
a novel architecture and algorithm. IEEE Transactions on cybernetics, 47(4), 946-959.

This project consist of deep multilayered ESN implementation having

1. Matlab Version
2. Python Version
3. Tensorflow Version

chmod +x ./run_train_test.sh
./run_train_test.sh

pip install ESN

from ESN.ML import ESN

        x = np.multiply((1 - a), x) + \
            np.multiply(a, np.tanh(np.add(np.multiply(W_L0, u), np.dot(W_reservoir_L0, x))))

L1 = ESN(W_L1, W_reservoir_L1, a)(x)

L2 = ESN(W_L2, W_reservoir_L2, a)(L1)

.
.
.
.

Ln = ESN(W_Ln, W_reservoir_Ln, a)(Ln-1)

Output of The Project on McKayGlass Timeseries Dataset

1. The weights everytime are initialized randomly but on scale recommendation is to optimize, tune and fix the initialization of both internal, external and reservoirs weights like any other networks except the readouts.