How is the mini-batch loss/error different from training or test loss/error?

[prosopher]

As far as I know, it is uncommon to describe evaluation metrics in terms of mini-batch

[songhwanjun]

The training loss and error are computed at the end of each epoch without any network update similar to validation ones. However, the mini-batch loss and error are accumulated all the iterations while continuously updating the model.