From the course: Neural Networks and Convolutional Neural Networks Essential Training

Unlock the full course today

Join today to access over 22,400 courses taught by industry experts or purchase this course individually.

Enhancements to CNNs

Enhancements to CNNs

From the course: Neural Networks and Convolutional Neural Networks Essential Training

Start my 1-month free trial

Enhancements to CNNs

- [Instructor] There are several enhancements to CNNs. But we'll just focus on two here. The purpose of Dropout is to prevent over fitting. The effect of this random Dropout forces the network to learn a redundant representation of the data. So Dropout is randomly killing each neuron in a layer of a training set with probability p. This Dropout probability is determined by you. And the common choice is p = 0.5 So this means that half of the neurons in a layer are dropped during training. Therefore the network cannot rely on an activation of any set of hidden units, since they may be turned off at any time during training. And the model is forced to learn more general and more robust patterns from the data. Remember that Dropout only takes place during training the data, and we don't use it when evaluating data. Image augmentation is the process of taking images that are already in a training dataset, and manipulating them to create many altered versions of the same image. This not…

Contents