1. In neurons, we introduce non-linearity to the result, z, by applying a function f() called the activation or transfer function. Refer section Artificial neurons.
2. Activation functions are used for introducing nonlinearity.
3. We calculate the gradient of the cost function with respect to the weights to minimize the error.
4. RNN predicts the output not only based on the current input but also on the previous hidden state.
5. While backpropagating the network if the gradient value becomes smaller and smaller it is called vanishing gradient problem if the gradient value becomes bigger then it is exploding gradient problem.
6. Gates are special structures in LSTM used to decide what information to keep, discard and update.
7. The pooling layer is used to reduce the dimensions of the feature maps and keeps only necessary details so that the amount of computation can be reduced.