Sequences and Attention

Five questions. Choose one answer for each, then submit to see your score and explanations.

The practice target is 80%. You can retry as often as you like, and every lesson stays accessible.

You can practice as a guest. Sign in before starting an attempt to save your scores.

1. What distinguishes a recurrent neural network when processing a sequence?A. It predicts every label from an unrelated random vectorB. It can only process a single scalar onceC. It updates a hidden state using current input and previous stateD. It must ignore the order of inputs

2. Why do LSTM and GRU architectures use gates?A. To shuffle the sequence into alphabetical orderB. To control what information is retained, updated, or exposedC. To guarantee unlimited perfect memoryD. To eliminate every trainable parameter

3. What does attention typically compute for a query?A. A weighted combination of value vectors based on relevance scoresB. Only the maximum raw input pixelC. The correct answer without learned representationsD. An unordered list of model filenames

4. A decoder predicts the next word during training. Which attention rule prevents it from looking at future words?A. Unrestricted access to every future positionB. Sorting all words by frequencyC. Removing all positional informationD. A causal mask

5. A self-attention model receives word vectors but no positional information. What important distinction can become difficult?A. Multiplying a matrix by another matrixB. Computing any attention weights at allC. Distinguishing sequences that contain the same words in different ordersD. Detecting that a vector has a numerical dimension

Review this module from its first lesson →

Sequences and Attention

Your result