What is backpropagation through time? Why is it useful in training RNNs$/$LSTMs$?$ Group of answer choices: $1)$ Backpropagation through time is a technique for updating weights in feedforward networks. It is useful in RNNs and LSTMs because it helps avoid overfitting by adjusting the learning rate based on time intervals. $2)$ Backpropagation through time is a method of unfolding RNNs over multiple time steps and applying backpropagation to each step. It is useful for training RNNs and LSTMs because it allows the network to capture long$-$term dependencies by computing gradients across time steps. $3)$ Backpropagation through time involves freezing the weights of the RNN or LSTM at each time step. It is useful because it prevents overfitting in recurrent networks by not allowing weight updates. $4)$ Backpropagation through time is a form of backpropagation that only applies to convolutional layers. It is useful for training RNNs and LSTMs because it ensures spatial features are preserved across layers.