Fig. 2.

With reduced learning rate of 0.001, there is fast convergence using pre-trained weights on the AlexNet model. Both training and validation loss reduce over the course of training, with accuracy on the validation dataset reaching approximately 99% at 90 epochs