Validation accuracy that appears different from the graph in DeepNetworkDesigner

Hi all!

I've been using DeepNetworkDesigner and have noticed that the final validation accuracy and validation loss do not seem to match what I would read off the plot. For instance, in the image below I would have expected training validation to be about 90% and the loss to be ~2. Instead, it reported 48% and a loss of nearly 8. Can anyone explain to me why this is happening and it is shooting up at the end?

Answer 1

[Jack Xiao]

it is overfit. add more training data, or add dropout layer, or reduce net layers.