The figure 2 is formed by the following programming: 
	1. fcn_ntk_init_prior_and_alignment_computation is for finding the PAC-Bayesian alignment under the fully connected network prior to considering the rho and kl_penalty. Note, the different proportion of data for prior training is conducted.
	2. cnn_ntk_init_prior_and_alignment_computation is for finding the PAC-Bayesian alignment under the convolutional neural network prior to considering the rho and kl_penalty. Note, the different proportion of data for prior training is conducted.
	3. fcn_ntk_init_posterior is a brutal-force grid search finding the generalization bound under fully connected network with MNIST. 
	4. cnn_ntk_init_posterior is a brutal-force grid search finding the generalization bound under convolutional neural network with CIFAR10. 
	5. fcn_correlation_drawing is for the first three rows in figure 2. In addition, the true values of PAC-Bayesian alignment is computed with the rho and kl_penalty. 
	6. cnn_correlation_drawing is for the first three rows in figure 2. In addition, the true values of PAC-Bayesian alignment is computed with the rho and kl_penalty. 
