Attention and Transformers

2 min

Objective: understand Query, Key, Value, multi-head attention and Transformer blocks.

Attention lets every position weight the information coming from other positions. From a representation X, we compute:

text
Q = XW_Q,  K = XW_K,  V = XW_V
Attention(Q,K,V) = softmax(QKᵀ / √d_k) V

The query Q describes what a position is looking for, the keys K what each position offers, the values V the information to aggregate. Dividing by √d_k avoids scores too large that would saturate softmax.

Multi-head attention

Several heads project the data into different subspaces, apply attention in parallel, concatenate their outputs, then reproject them. A head is not guaranteed to correspond to a stable human concept.

Preview — the rest of the lesson is for enrolled readers.

Already enrolled with a code?

Your access is tied to your account, not to this link. Sign in with the same email you used in class: your course is waiting, no need to enter the code again.

Sign inNo account yet? Create one
This lesson is part of the “Specialized Architectures” module

The first modules of the course are open to everyone. For the rest you have three options: buy this course once and for all, subscribe, or enter the code handed out in class.

Are you a student on this course?

The code is tied to your account: sign in or create an account and it will be applied automatically when you come back.

No account yet? Create one