Accuracy increases, but Test loss does not decrease

[sia]

Hello. Could anyone tell me what is happening here? How to analyze this kind of Test loss?

[Brigit Schroeder]

You need to compare this to the train loss, plotting both on the same accuracy plot. In general, you want the train lost and test loss to increase at the same approximate rate.

[Przemek D]

@Brigit: loss and accuracy are two entirely different things.
Regarding sia's plots: accuracy increasing with no decrease of loss really looks weird. Are you sure your plotting code is okay, and you're not making some simple mistake like plotting training accuracy with test loss (in this case you could guess that the net is overfitting)?