Self-Check Quiz: Convolutional Architectures and Beyond

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

Question 1 of 5

What is the main advantage of weight reuse in convolutional neural networks?

It dramatically reduces the number of parameters while enabling massive parallelism.It enables each neuron to process a different part of the input, increasing model capacity.It eliminates the need for handcrafted filters by using max pooling.It allows the network to adapt its filter sizes dynamically during training.

Question 2 of 5

What is the primary purpose of label smoothing as introduced in Inception v3?

It prevents overconfident predictions and improves generalization.It encourages logits to grow without bound for better class separation.It adds noise to the labels to make the training task harder.It uses a separate network to predict label distributions.

Question 3 of 5

How does the skip connection in a ResNet unit help mitigate the vanishing gradient problem?

It provides a direct gradient path to earlier layers.It reduces the number of parameters by using 1x1 convolutions.It normalizes the activations within each mini-batch.It splits the input into multiple groups for parallel processing.

Question 4 of 5

What does gradient clipping do to handle exploding gradients in RNNs?

It rescales the gradient vector to limit its norm when it exceeds a threshold.It randomly drops a fraction of gradient elements to reduce explosion risk.It automatically reduces the learning rate when gradients become large.It applies a fixed scaling factor to all gradients after backpropagation.

Question 5 of 5

How does the constant error carousel in an LSTM help preserve gradients over long sequences?

It provides a path where the gradient can flow unchanged across time steps.It uses a reset gate to control the flow of gradients into the cell.It initializes the cell state with large values to prevent decay.It applies layer normalization to the cell state before each update.

0 / 5 answered

Generated from lecture recordings by the autosummary pipeline.