Question 25
Domain 2: Core Machine Learning, AI, and Transformer FoundationsIn the self-attention mechanism, which matrix represents the token seeking relevant information?
Correct answer: C
Explanation
In self-attention, the Query (Q) matrix represents the token that is "seeking" information from other tokens. It is compared with Keys (K) to determine which values (V) should be weighted most heavily, so Q drives the lookup for relevant context.
Why each option is right or wrong
A. Key (K)
B. Value (V)
C. Query (Q)
In scaled dot-product self-attention, the input is linearly projected into Query (Q), Key (K), and Value (V) matrices, and the attention scores are computed from Q·K^T before being normalized with softmax. The Q matrix is the one used to probe the sequence for matching keys, so it is the token representation performing the lookup for context; K identifies what each token offers, and V carries the content that is ultimately aggregated.
D. Output (O)