I’ve experimented with a variety of network architectures and have gone to 50 epochs, but cannot seem to get training accuracy to reach anything above 91%. I’ve used anywhere between 3 and 5 convolutional layers with a variety of filters each. All activation functions are relu, except the last one (sigmoid). I’ve tried different both Adam and RMS optimizers. I have not used any augmentation-- as I saw that is for next week, and would prevent overtraining. Can I get some suggestions on how to achieve the desired training accuracy? My validation accuracy is plenty high.
Did you use a max pooling layer after every conv layer?
Hi I had same difficulty too but I have achieved this with 3 convolutional with max pool layer and 2 hidden layer and I also set 100 epochs with callback function accuracy of 95 % its working fine.
maybe try to set a lower batch_size value~