Skip-gram and CBOW
Shallow word2vec architectures that learn word embeddings from context: CBOW predicts a target word from neighbors; skip-gram predicts neighbors from a target word.
Skip-gram and CBOW (CBOW) are the two main training modes in word2vec. CBOW predicts a masked word from surrounding context tokens. Skip-gram reverses the direction: given one word, it predicts multiple nearby words. Both learn dense word embeddings from large unlabeled corpora (article).