In this session, you leant to build and train CNNs Keras and experimented some hyperparameters of the model. You also practised manually computing the number of parameters ,output size etc. of CNN-based architectures.
Based on these experiments, we saw that the performance of CNNs depends heavily on multiple hyperparameters – the number of layers, number feature maps in each layers,the use of dropouts, batch normalisation, etc. Thus, it is advisable to first fine-tune your model hyperparameters by conducting lots of experiments . Only when you are convinced that you have found the right set of hyperparameters you should train the model with a lager number of epochs ( since almost always the amounts of time and computing power you have is limited)
In the next session, you will study the architectures of some popular deep convolutional networks, learn to train CNNs in Python + Keras, and use large pre-trained networks for your own tasks using transfer learning.
Report an error