Table 9 Performance boost of GP-PINN on the Helmholtz benchmarks using transfer learning
From: Transfer learning for deep neural network-based partial differential equations solving
Subtask | Original | Transfer | Boost |
|---|---|---|---|
case 2 | 9.01e-03 | 6.51e-03 | 27.7% |
case 3 | 3.48e-03 | 3.30e-03 | 5.2% |
case 4 | 9.20e-03 | 6.14e-03 | 33.3% |
case 5 | 2.33e-02 | 1.89e-02 | 18.9% |
case 6 | 3.04e-02 | 2.75e-02 | 9.5% |
case 7 | 7.28e-02 | 3.71e-02 | 49.0% |
case 8 | 8.59e-01 | 8.23e-02 | 90.4% |