Data Science and Machine Learning (Theory and Projects) A to Z - Vanishing Gradients in RNN: GRU

What is a primary challenge in recurrent neural networks that affects their performance?

• Overfitting

• High computational cost

• Data scarcity

• Vanishing gradient

Which of the following is a more recent solution to the vanishing gradient problem?

• LSTM

• CNN

• GRU

• RNN

What is the main advantage of GRU over LSTM?

• Higher accuracy

• Simplicity

• Faster training

• Better generalization

In a simple recurrent neural network, which activation function is often preferred?

• Sigmoid

• ReLU

• Tanh

• Softmax

Why is the tanh activation function preferred in RNNs?

• It is easier to implement

• It is less prone to overfitting

• It maintains memory better

• It is computationally efficient

What is the role of the update gate in a GRU?

• To decide whether to update activations

• To compute loss

• To normalize inputs

• To initialize weights

In GRU, what happens if the update gate is set to zero?

• The previous activation is carried forward

• The activation is reset

• The candidate activation is used

• The model stops training

How does GRU handle long-term dependencies compared to simple RNNs?

• By using more layers

• By using update gates

• By reducing model size

• By increasing learning rate

What makes GRU a state-of-the-art solution for vanishing gradient problems?

• Its low computational cost

• Its use of gates to manage memory

• Its ability to handle large datasets

• Its compatibility with other models

What will be discussed in the next video following the GRU?

• Data preprocessing

• Convolutional Neural Networks

• LSTM

• Advanced GRU techniques