Table 2 Ablation studies examine the impact of various neural solver combinations on model performance.

From: Physics-informed neural networks with hybrid Kolmogorov-Arnold network and augmented Lagrangian function for solving partial differential equations

Neural solvers

Relative \(L^{2}\) error

MSE

MLP+MLP

\({1.93\times 10^{-2}}\)

\({2.87\times 10^{-4}}\)

GRU+GRU

\({7.03\times 10^{-3}}\)

\({8.39\times 10^{-5}}\)

GRU+MLP

\({{8.51\times 10^{-3}}}\)

\({{8.86\times 10^{-5}}}\)

KAN+KAN

\({4.71\times 10^{-3}}\)

\({5.48\times 10^{-6}}\)

GRU+KAN

\({5.14\times 10^{-4}}\)

\({2.25\times 10^{-7}}\)

Back to article page