A time$-$series input is scanned for a pattern using the following MLP$.$ While scanning, the

entire network strides two time steps at a time. Subsequently the outputs of the MLP at each

time step are combined through a softmax. In the figure, neurons of identical color within

each layer have identical responses.

The same operation can be performed using a time$-$delay neural network. What will the

architecture of this TDNN be$?$

Hint: Piazza @$659$

Three convolutional layers. $3$ filters of kernel$-$width $2$ in the first layer, with stride of $2,3$ filters of

kernel$-$width $6$ with stride $3$ in the second layer, and $2$ filters of kernel$-$width $6$ with stride $3$ in layer $3.$

Three convolutional layers. $3$ filters of kernel$-$width $2$ in the first layer, with stride $2,3$ filters of kernel$-$

width $2$ in the second layer, with stride $1,$ and $2$ filters of kernel$-$width $2$ in the third layer, with a stride

of $1.$

Three convolutional layers. $9$ filters of kernel$-$width $2$ in the first layer, with stride $2,6$ filters of kernel$-$

width $2$ in the second layer, with stride $2,$ and $2$ filters of kernel$-$width $2$ in the final layer with stride $1.$

Two convolutional layers and a flat layer. $9$ filters of kernel$-$width $2$ in the first layer, with stride $2\mathrm{.}6$

filters of kernel$-$width $2$ with stride $3$ in the second layer. The final flat layer has two neurons with all

the outputs of the second layer going to them.