Table 4 The table provides the resulting MSE for our deep neural network models

From: Solving Boltzmann optimization problems with deep learning

Problem

(N, n, α)

Size of training data

Size of testing data

MSE

1

(9, 4, 1)

51,773

12,944

0.000262

2

(11, 5, 1)

126,428

35,767

0.000910

3

(14, 6, 2)

7636

1910

0.0206542

4

(15, 6, 3)

7477

1870

0.0203343

  1. The neural networks, created using DJINN framework, were initialized with a random forest regressor with only three trees and a maximum depth of ten; whereas our random forest regressors had 100 trees with larger depth. Consequently, the number of parameters required is far fewer than that required for the random forest regression model and yet, the MSE of the results are similar.
Back to article page