ML Chapter 05

What is a computational graph used for in neural networks?

• To display images
• To store data
• To visualize and compute operations step by step
• To compress weights

In a computational graph, what do the edges represent?

• Loops
• Data inputs
• The result of local computations
• Error rates

What is the chain rule used for in backpropagation?

• To skip intermediate steps
• To calculate forward passes
• To propagate gradients through layers
• To normalize inputs

What direction does backpropagation travel through the computational graph?

• Left to right
• Random
• Right to left
• Center to edges

What does the MulLayer represent in a computational graph?

• A layer that adds inputs
• A multiplication node
• A loss function
• A softmax operation

What is the purpose of the AddLayer in the backpropagation system?

• It filters values
• It performs subtraction
• It adds values and propagates the gradient unchanged
• It stores weights

What happens in the ReLU activation during backpropagation?

• All gradients are reversed
• Only positive inputs propagate gradients
• Gradients are squared
• It turns into a sigmoid

What is the Affine layer responsible for in a neural network?

• Applying dropout
• Performing element-wise activation
• Matrix multiplication and adding bias
• Converting to one-hot vectors

During backpropagation in the Affine layer, what is transposed?

• The input X
• The bias B
• The gradient dL/dW
• The output Y

What is the main reason for using mini-batches during training?

• To overfit better
• To reduce loss permanently
• To improve training speed and memory efficiency
• To reset parameters