Optimisation issue

[colibrisson]

Today I tried to train a recognition model, but despite multiple attempts with different hyperparameters, training failed to converge: stagnating training_loss and null val_accuracy. I tried with another dataset that I used to train a recognition model back in April but, despite using the same parameters as before, training also did not converge. I haven't kept up with the latest changes to kraken so I'm not sure what might be causing the issue.