An Efficient Evolutionary Neural Architecture Search Algorithm Without Training

. 2025 Jun 29;10(7):421. doi: 10.3390/biomimetics10070421

Algorithm 1 Neighborhood competition operations

Require:
$A_{i, j} \in R^{n}$ : individual at position $(i, j)$ ;

$A_{i, j}^{neighbors} \subset R^{n}$ : neighbor individuals;

$P_{s} \in [0, 1]$ : selection probability
Ensure:
$A_{i, j}^{'} \in R^{n}$ : updated individual at position $(i, j)$
1:
$M_{i, j}$ ← None
2:
for each A in $A_{i, j}^{n e i g h b o r s}$ do
3:
if $Fitness (A)$ > $Fitness (M_{i, j})$ then
4:
$M_{i, j}$ ←A
5:
end if
6:
end for
7:
if $Fitness (A_{i, j}) < Fitness (M_{i, j})$ then
8:
if $Rand (0, 1)$ < $P_{s}$ then
9:
$N e w I n d i v i d u a l$ ← Strategy1( )
10:
▹ Use Strategy 1 to generate new individual
11:
else
12:
$N e w I n d i v i d u a l$ ← Strategy2( )
13:
▹ Use Strategy 2 to generate new individual
14:
end if
15:
else
16:
$N e w I n d i v i d u a l$ $\leftarrow A_{i, j}$ ▹ Retain the original individual
17:
end if
18:
Return $N e w I n d i v i d u a l$ as the updated individual $A_{i, j}$