Algorithm 7: Proposed ETLBOCBL-CNN
Inputs: $N$, $D$, ${ℛ}^{train}$, ${ℛ}^{valid}$, $S^{batch}$, ${\epsilon }^{train}$, ${\epsilon }^{FT}$, $R^L$, $C^{num}$, $N_{min}^{Conv}$, $N_{max}^{Conv}$, $N_{min}^{Fil}$, $N_{max}^{Fil}$, $S_{min}^{Ker}$, $S_{max}^{Ker}$, $S_{min}^{Pool}$, $S_{max}^{Pool}$, $S_{min}^{Str}$, $S_{max}^{Str}$, $N_{min}^{FC}$, $N_{max}^{FC}$, $N_{min}^{Neu}$, $N_{max}^{Neu}$, $S^{Group}$, $G$
01:	Load ${ℛ}^{train}$ and ${ℛ}^{valid}$ from the directory;
02:	Initialize the population $P=\left[X_1,\dots ,X_n,\dots ,X_N\right]$ using Algorithm 1;
03:	Initialize the iteration counter as $t\leftarrow 0$;
04:	while $t<T^{max}$ do
05:	Generate $P^{off}$ and update $X^{Teacher}$ using modified teacher phase (Algorithm 4);
06:	Update $P^{off}$ and $X^{Teacher}$ using modified learner phase (Algorithm 5);
07:	Determine $P^{Next}$ using tri-criterion selection scheme (Algorithm 6);
08:	$P\leftarrow P^{Next} ;$
09:	$t\leftarrow t+1$;
10:	end while
11:	Fully train the CNN architecture constructed from $X^{Teacher}.Pos$ with larger ${\epsilon }^{FT}$ (Algorithm 2);
Output: $X^{Teacher}$ and its corresponding optimal CNN architecture