Optimizing Image Classification: Automated Deep Learning Architecture Crafting with Network and Learning Hyperparameter Tuning

. 2023 Nov 4;8(7):525. doi: 10.3390/biomimetics8070525

Algorithm 1: Population Initialization of ETLBOCBL-CNN
Input: N, $N_{m i n}^{C o n v}$ , $N_{m a x}^{C o n v}$ , $N_{m i n}^{F i l}$ , $N_{m a x}^{F i l}$ , $S_{m i n}^{K e r}$ , $S_{m a x}^{K e r}$ , $S_{m i n}^{P o o l}$ , $S_{m a x}^{P o o l}$ , $S_{m i n}^{S t r}$ , $S_{m a x}^{S t r}$ , $N_{m i n}^{F C}$ , $N_{m a x}^{F C}$ , $N_{m i n}^{N e u}$ , $N_{m a x}^{N e u}$ , $L H_{m i n}^{O p t}$ , $L H_{m i n}^{L R}$ , $L H_{m i n}^{I n t},$ $L H_{m i n}^{L 2}$ , $L H_{m a x}^{O p t}$ , $L H_{m a x}^{L R}$ , $L H_{m a x}^{I n t}$ , $L H_{m a x}^{L 2}$
01:	Compute the dimensional size as $D = 5 N_{m a x}^{C o n v} + N_{m a x}^{F C} + 6$ ;
02:	Initialize teacher solution as $X^{T e a c h e r} . P o s \leftarrow \emptyset$ and $X^{T e a c h e r} . E r r \leftarrow \infty$ ;
03:	for n = 1 to N do
04:	Initialize $X_{n} . P o s \leftarrow \emptyset$ ;
05:	for d = 1 to D do
06:	if $d = = 1$ then
07:	Assign $X_{n} . {P o s}_{d}$ with $N^{C o n v} \in {N_{m i n}^{C o n v}, N_{m a x}^{C o n v}}$ ;
08:	else if $d = = 2 l$ then
09:	Assign $X_{n} . {P o s}_{d}$ with $N_{l}^{F i l} \in {N_{m i n}^{F i l}, N_{m a x}^{F i l}}$ for $l = 1, \dots, N_{m a x}^{C o n v}$ ;
10:	else if $d = = 2 l + 1$ then
11:	Assign $X_{n} . {P o s}_{d}$ with $S_{l}^{K e r} \in {S_{m i n}^{K e r}, S_{m a x}^{K e r}}$ for $l = 1, \dots, N_{m a x}^{C o n v}$ ;
12:	else if $d = = 2 N_{m a x}^{C o n v} + 3 l - 1$ then
13:	Assign $X_{n} . {P o s}_{d}$ with $P_{l}^{P o o l} \in [0, 1]$ for $l = 1, \dots, N_{m a x}^{C o n v}$ ;
14:	else if $d = = 2 N_{m a x}^{C o n v} + 3 l$ then
15:	Assign $X_{n} . {P o s}_{d}$ with $S_{l}^{P o o l} \in {S_{m i n}^{P o o l}, S_{m a x}^{P o o l}}$ for $l = 1, \dots, N_{m a x}^{C o n v}$ ;
16:	else if $d = = 2 N_{m a x}^{C o n v} + 3 l + 1$ then
17:	Assign $X_{n} . {P o s}_{d}$ with $S_{l}^{S t r} \in {S_{m i n}^{S t r}, S_{m a x}^{S t r}}$ for $l = 1, \dots, N_{m a x}^{C o n v}$ ;
18:	else if $d = = 5 N_{m a x}^{C o n v} + 2$ then
19:	Assign $X_{n} . {P o s}_{d}$ with $N^{F C} \in {N_{m i n}^{F C}, N_{m a x}^{F C}}$ ;
20:	else if $d = = (5 N_{m a x}^{C o n v} + 2) + q$ then
21:	Assign $X_{n} . {P o s}_{d}$ with $N_{q}^{N e u} \in {N_{m i n}^{N e u}, N_{m a x}^{N e u}}$ for $q = 1, \dots, N_{m a x}^{F C}$ ;
22:	else if $d = = 5 N_{m a x}^{C o n v} + N_{m a x}^{F C} + 3$ then
23:	Assign $X_{n} . {P o s}_{d}$ with $L H^{O p t} \in {L H_{m i n}^{O p t}, L H_{m a x}^{O p t}}$ ;
24:	else if $d = = 5 N_{m a x}^{C o n v} + N_{m a x}^{F C} + 4$ then
25:	Assign $X_{n} . {P o s}_{d}$ with $L H^{L R} \in {L H_{m i n}^{L R}, L H_{m a x}^{L R}}$ ;
26:	else if $d = = 5 N_{m a x}^{C o n v} + N_{m a x}^{F C} + 5$ then
27:	Assign $X_{n} . {P o s}_{d}$ with $L H^{I n t} \in {L H_{m i n}^{I n t}, L H_{m a x}^{I n t}}$ ;
28:	else if $d = = 5 N_{m a x}^{C o n v} + N_{m a x}^{F C} + 6$ then
29:	Assign $X_{n} . {P o s}_{d}$ with $L H^{L 2} \in {L H_{m i n}^{L 2}, L H_{m a x}^{L 2}}$ ;
30:	end if
31:	end for
32:	Fitness evaluation of $X_{n} . P o s$ as $X_{n} . E r$ using Algorithm 2;
33:	if $X_{n} . E r r < X^{T e a c h e r} . E r r$ then
34:	$X^{T e a c h e r} . P o s \leftarrow X_{n} . P o s$ , $X^{T e a c h e r} . E r r \leftarrow X_{n} . E r r$ ;
35:	end if
36:	end for
Output: $P = [X_{1}, \dots, X_{n}, \dots ., X_{N}]$ , $X^{T e a c h e r}$