training help

[Eric J]

Hi, Thanks very much for reading my question.

I am training a network.

I change the learning rate in 0.01,0.001, 0.0001, 0.00001 .

The training curve and test curve is shown below.

I am using a batch size 7 since I have low GPU memory.

It seems the train loss will decrease to 0 while the test loss for

all the learning rate will get plateau.

I do not know what happened?  Is it because overfitting?

I do appreciate your help if you could share me your experiences 

in training especially how to analysis the learning curve.

Thanks very much!