The Elman Network $($without output activation function and bias units$)$ can be defined as

$s(t)=Wx(t)+a(t-1)$

$a(t)=f(s(t))$

hat$(y)(t)=Va(t)$

Verbleibende Zeit $0$:$49$:$44$

with input vectors $x(t),$ hidden pre$-$activation vectors $s(t),$ hidden activation vectors $a(t),$ activation function $f(*),$ and parameter matrices $W,V.$ Which of the following statements are true?

a$.$ The recurrent weights are equal to the identity matrix.

b$.$ The Elman network is well suited for training with RTRL because the recurrent Jacobian has a diagonal structure and this reduces the computational complexity.

c$.$ The matrix $W$ contains the recurrent weights.

d$.$ The hidden units of the Elman network are interconnected, i$.$e$.,$ at time $t$ each hidden unit receives input from all other hidden units at time $t-1.$

e$.$ The recurrent Jacobian of the hidden activations is $\frac{\text{d}\text{e}\text{l}\text{a}(t)}{\text{d}\text{e}\text{l}\text{a}(t-1)}=$diag$(f^{\text{'}}(s(t))).$