Keras cifar10 test example and test loss is lower than training loss

Question

I play with Keras cifar10 example, which you can find here here . I recreated the model (i.e. not the same file, but everything else is almost the same) and you can find it here .

The model is identical, and I train the model for 30 eras with 0.2 validation validation for 50,000 training images. I can not understand the result that I get. My reliability and testing loss is less than less training (back, the accuracy of training is lower compared to the accuracy of testing and testing):

                      Loss       Accuracy
   Training          1.345          0.572
 Validation          1.184          0.596
       Test           1.19          0.596

Looking at the plot, I'm not sure why the error in learning is starting to increase again so much. Do I need to reduce the number of eras that I train or, perhaps, introduce an early stop? Can another model architecture help? If so, what are some good suggestions?

Thank.

Answer 1

This is a rare occurrence, but it does occur from time to time. There are several reasons why this may be so:

• , , , , , .
• - - .