Hi,
Since the validation loss is not changing a lot so the model is in a plateau. Either it reached the maximum accuracy that can be achieved, or it is in local minima. I suggest checking it with higher learning rate at later epochs so that if the model is in local minimum it can come out of it. You can schedule the learning rate in trainingOptions with ‘LearnRateDropPeriod’ and ‘LearnRateDropFactor’ if using trainNetwork.
