17. Transformers Explained Easily: Part 1 - Generative Music AI

Learn the intuition, theory, and mathematical formalization of transformer architectures, Transformers are widely used in Deep Learning. They now dominate Natural Language Processing, computer vision, and are extensively used in audio and music processing.

Get the lecture slides:
https://github.com/musikalkemist/gene...

Website of the Generative Music AI Workshop in Barcelona:
https://www.upf.edu/web/mtg/generativ...

Sign up to The Sound of AI Slack Community to join the discussion:
https://valeriovelardo.com/the-sound-...

======================================

Interested in music AI consulting?
https://thesoundofai.com/consulting.html

Interested in music AI recruitment?
https://thesoundofai.com/recruitment....

Become a Python ninja with my Advanced Python Programming course:
https://the-sound-of-ai-academy.teach...

Connect with Valerio on LinkedIn:
  / valeriovelardo  

Follow Valerio on Twitter:
  / musikalkemist  

======================================

Content
0:00 Intro
2:57 Context
6:01 The intuition
8:30 Encoder
12:15 Encoder block
12:42 Self-attention
16:36 Matrices
17:04 Input matrix
18:29 Query, key, value matrices
21:20 Self-attention formula
22:01 Self-attention: Step 1
28:07 Self-attention: Step 2
30:02 Self-attention: Step 3
32:59 Self-attention: Step 4
45:45 Self-attention: Visual recap
47:47 Multi-head attention
50:49 The propblem of sequence order
52:33 Positional encoding
55:30 How to compute positional encoding
1:02:01 Feedforward layer
1:03:27 Add & norm layer
1:06:09 Deeper meaning of encoder components
1:07:10 Encoder step-by-step
1:10:37 Key takeaways
1:12:20 What next?