Query, Key, and Value in Attention

What are Query, Key, and Value in attention?

  First visit stackexchange [1] to get familiar with what issue we are dealing with here. After you read all of the things written in the page, you would probably not clear about the issue. But, as suggested in the page, pay a visit to this video clip about SVD [2] to have some idea about "compatibility function". Intuitively and naively speaking, it would feel like casting some two seemlingly unsimilar vectors to find some common grounds. (Just as the example illustrates, two people at the outlook whose ratings are not similar at all, after all, have commonality to like sci-fi movies.)

 Refer to [3] for details. If you think you know what's going on by looking at below figures, then you are good to go.

[1] neural networks - What exactly are keys, queries, and values in attention mechanisms? - Cross Validated (stackexchange.com)

[2] https://youtu.be/K38wVcdNuFc?t=10

[3] Deep Learning: The Transformer. Sequence-to-Sequence (Seq2Seq) models… | by Mohammed Terry-Jack | Medium