#fasttext

Question 1: Single choice

mediumAI-checkedText Representation & Embeddings

fastText turns skip-gram into a subword model by replacing the input-side vector, the center word's $\mathbf{v}_w$ , with the sum of its character n-gram vectors. Applied to CBOW, which predicts the center word from the average of its context-word vectors, which vectors become n-gram sums?

Each context word's input vector, summed from its n-grams and then averaged as usual1
Only the predicted center word's output vector; context words keep whole-word vectors2
The n-gram vectors of every word in the vocabulary, averaged into a single shared vector3
The context words' n-gram vectors, concatenated instead of averaged4

Timed test