MatConvNet, Error vs Objective Plot Understanding

Question

How is one supposed to interpret the graph output for the DAG wrapper? There is absolutely minimum documentation on this. What does energy vs epochs mean when relating to "Objective"? What is the difference between the ERROR plot and the OBJECTIVE plot? Does MatConvNet try to minimize objective and not the error plot? How does MatConvNet handle these output graphs in DAG wrapper vs Simple Wrapper? Please, someone has to have some insight to this. I've searched for responses, but all that was given was unreliable assumptions...

Attaching example images below.

Output Plot #1 (Potential Overfitting)

Output Plot #2 (Error and Training convergence)

Answer 1

Like other deep learning frameworks, networks try to minimize "objective". The objective function is designed such that it is very correlated with the "error". You can't directly optimize the error since it's discrete. You have to find some proxy of it to optimize it.

How does MatConvNet handle these output graphs in DAG wrapper vs Simple Wrapper? If I remember correctly, MatConvNet looks for variables in DAG that has the name "error" and "loss" and display them.

Let me know if you have other questions.