Table 1 Network and hyper parameter settings used in the experiments
From: Geometry aware physics informed neural network surrogate for solving Navier–Stokes equation (GAPINN)
SEN | PINN | BCN | |
|---|---|---|---|
Network parameter | |||
Layers | 4 × 1d Convolutions (feature size: 128, 128, 256, 512) FCN (3 hidden layer with 256 neurons, nk = 60); \(\beta = 0.001\) | 2 hidden layer (each 256 neurons) | 2 hidden layer (each 25 neurons) |
Training parameter | |||
Batch size | 50 | 10 | 10 |
Optimizer | Adam [16] Betas = (0.9, 0.999) No weight decay No amsgrad | Adam Betas = (0.9, 0.999) No weight decay No amsgrad | Adam Betas = (0.9, 0.999) No weight decay No amsgrad |
Learning rate | 1e−3 | 1e−3 | 1e−3 |
Epochs | 650 | 1300 | 3000 |