Quiz: Recurrent Neural Networks: Processing Sequences and Time

Question 1 of 7

What type of data does a CNN (Convolutional Neural Network) excel at processing?

Images where spatial relationships matter

Text where temporal order matters

Speech where phonemes combine over time

Time series data with sequential dependencies

Question 2 of 7

What two inputs does an RNN cell receive at each step?

The current token and the hidden state from the previous step

The current token and the cell state

The input gate and the forget gate

The current word and the output from the next step

Question 3 of 7

What causes the vanishing gradient problem in basic RNNs?

Gradients are multiplied by weights at each step, causing them to shrink exponentially toward zero

The hidden state becomes too large to process

The network runs out of memory

The learning rate decreases over time

Question 4 of 7

What are the two types of state that LSTMs maintain?

Cell state for long-term memory and hidden state for immediate computations

Input state and output state

Forward state and backward state

Training state and prediction state

Question 5 of 7

What does the forget gate in an LSTM decide?

What to discard from cell state

What new information to store

What from cell state to output at this step

How much past information to keep

Question 6 of 7

How many gates do GRUs use compared to LSTMs?

Two gates (reset and update) instead of three

Three gates like LSTMs

Four gates for better performance

One combined gate

Question 7 of 7

Why are Transformers 100x faster to train than RNNs?

They process entire sequences in parallel through attention mechanisms instead of sequentially

They use fewer parameters

They don't require backpropagation

They only work with shorter sequences