Algorithm 1 HARNAS Search Algorithm for Neural Architectures

Input: Search space (e.g., hyperparameters including layers and blocks), fitness value (F), $n_{population}$, $n_{epochs}$   Output: Best identifying network ($IN$${}^{*}$) of the Pareto optimal set. 1:Initialize the population networks (PNs); 2:for each $i\in [1,n_{population}]$ do 3:  for each $j\in [1,PNs]$ do 4:    Train $I{N}_j$ for $n_{epochs}$; 5:  end for 6:  Update $PNs$ with F with the EAs; 7:end for 8:return${\left(PN\right)}_i$=$n_{generations}$