Dimension réseau de neurones

Bonjour
J'ai vu le QCM suivant.

Suppose we want to train an MLP to perform a multi-class classification ("cat", "dog", "horse") of $10 x 10$ input images.
The MLP has two hidden layers of 20 units each.
What are the parameters to be trained?
(only one possibility)
Select one:
a. The matrices $\mathrm{W}_{-} 1$ of dimension $(100,20)$, W_2 of dimension $(20,20), \mathrm{W}_{-} 3$ of dimension $(20,3)$ and the bias b_1 of dimension (100), b_2 of dimension (20), b_3 of dimension (20)
b. The matrices $\mathrm{W}_{-} 1$ of dimension $(100,20)$, W_2 of dimension $(20,20)$, W_3 of dimension (20,1) and the bias b_1 of dimension (20), b_2 of dimension (20), b_3 of dimension (1)
c. The matrices W_1 of dimension (100,20), W_2 of dimension (20,20), and the bias b_1 of dimension (20), b_2 of dimension (20)
d. The matrices $\mathrm{W}_{-} 1$ of dimension $(100,20)$, W_2 of dimension $(20,20), \mathrm{W}_{-} 3$ of dimension $(20,3)$ and the bias b_1 of dimension (20), b_2 of dimension (20), b_3 of dimension (3) $\mathfrak{V}$


Et pour moi aucun n'est juste, pour moi c'est la d. mais avec $(20,100)$ et non $(100,20)$ je ne comprends pas mon erreur.