Why are RNNs susceptible to issues with their gradients?1 pointGradients can grow exponentiallyGradients can quickly drop and stabilize at near zeroPropagation of errors due to the recurrent characteristicNumerical computation of gradients can drive into instabilitiesAll of the above

Recurrent Neural Networks (RNNs) are susceptible to issues with their gradients due to several reasons:

1. Gradients can grow exponentially: This is known as the exploding gradients problem. In this case, the values of a gradient can become so large that they cause an overflow (i.e., values that are too large to be represented in the computer's memory), which can lead to unstable models.

2. Gradients can quickly drop and stabilize at near zero: This is known as the vanishing gradients problem. When the gradient is very small, the weights of the network are almost not updated during the backpropagation process, which means that the network stops learning or learns very slowly.

3. Propagation of errors due to the recurrent characteristic: The recurrent nature of RNNs means that the output at time step t is dependent on the computations performed at all previous time steps. This means that the error gradients that are backpropagated can explode or vanish, depending on the nature of the activation function and the weight initialization scheme.

4. Numerical computation of gradients can drive into instabilities: The process of computing gradients involves performing operations on small numbers, which can lead to numerical instability. This can result in gradients that are either very large