2.4.3 transpose confusion ∇x (Ax) = A^T

d2l-ai / d2l-en

Interactive deep learning book with multi-framework code, math, and discussions. Adopted at 500 universities from 70 countries including Stanford, MIT, Harvard, and Cambridge.

https://D2L.ai

Other

22.62k stars 4.2k forks source link

2.4.3 transpose confusion ∇x (Ax) = A^T #2532

Open PilotLeoYan opened 1 year ago

PilotLeoYan commented 1 year ago

2.4.3 Partial Derivatives and Gradients

For all A ∈ R ^ (m×n) we have ∇x (Ax) = A^⊤ ? The correct gradient is ∇x (Ax) = A https://www.youtube.com/watch?v=e73033jZTCI https://www.matrixcalculus.org/