Table 5.

Hyperparameter configurations for training and testing the overall methodology.

Processes	Hyperparameters	Parameters setting
ExPSO	Exponential weight parameter	$a=2$
	Cognitive acceleration best-coefficient	$b=2$
	Social acceleration best-coefficient	$c=2$
	Cognitive acceleration worst-coefficient	$d=-1$
	Social acceleration worst-coefficient	$e=-1$
	Cognitive scaling parameter	$c_1=-1$
	Social scaling parameter	$c_2=2$
	Inertia weight	$W=0.9$
	Damping coefficient	$r=0.9$
	Exploration number of iterations	$t=10$
	Decreasing velocity coefficient	$k=0.2$
	Number of particles of subpopulation 1,2, and 3	$N_1,N_2,N_3=10$
Compression	Learning rate (We maintain a fixed learning rate for our initial evaluation. However, for optimization problems with complex landscapes, we recommend starting with a fixed learning rate and subsequently considering updates to it as needed)	$Lr=3e-5$
	Epsilon	$Eps=1e-8$
	Compression dimensionally	$D=50257\ast 1024$
	Max iteration limit	$MaxIt=100$
	Early stopping patience	$20epochs$
	Epochs	$100epochs$
	Lower bounds	$Lb=-1$
	Upper bounds	$Ub=1$
Multi-expert	Number of heads	$H=5$
	Number of experts	$L=3$
XPSO	$n=0.2,{\mathit{Stag}}_{\mathit{max}}=5,p=0.5$
MSPSO	$X=0.7298,c_1=c_2=1.49445,R_2=10$
FMPSO	$c_H=1.2,c_m=0.7,c_l=0.2$
PPSO	Inertia weight	$w=0$