Convolutional Neural Networks

One answer per question, one attempt per quiz — you cannot retake it.

Question 1 of 5

Why can't the human brain perform gradient descent for learning, according to the lecture?

Biological neurons cannot store adjustable weights.Neurons can only compute forward, not the backward pass.The brain lacks a well-defined loss function to minimize.Feedback connections in the brain are too slow for gradient descent.

Question 2 of 5

What is the essential property that defines a convolutional layer?

It uses a small network applied to non-overlapping tiles.It applies the same transformation to every input tile.It connects each neuron to all neurons in the previous layer.It uses max pooling to reduce spatial dimensions.

Question 3 of 5

Why is max pooling generally preferred over average pooling in convolutional networks for vision?

Max pooling is computationally cheaper than average pooling.Most vision features are presence detectors, so max pooling works best.Average pooling introduces trainable parameters, while max pooling does not.Max pooling reduces spatial dimensions more aggressively than average pooling.

Question 4 of 5

What is the primary motivation for including pooling layers in a CNN, according to the lecture?

To introduce non-linearity between convolutional layers.To reduce spatial dimensions, increasing the receptive field.To increase the number of feature maps without adding parameters.To regularize the network by adding noise to the feature maps.

Question 5 of 5

Why do modern CNN architectures often prefer strided convolutions over separate pooling layers?

Strided convolutions introduce non-linearity, which pooling lacks.Strided convolutions keep gradient flow intact, unlike pooling.Strided convolutions have fewer parameters than pooling layers.Strided convolutions increase the receptive field more rapidly than pooling.

0 / 5 answered

Generated from lecture recordings by the autosummary pipeline.