From: Deep convolutional architectures for extrapolative forecasts in time-dependent flow problems
Hyperparameters | Values |
---|---|
Sequence length (\(n_t\)) | 5, 10, 20 |
CNN block channels | [50, 50], [100, 100], [200, 200] |
Latent dimension (m) | 12,25,50 |
Kernel size (k) | 3, 5, 7, 9 |
Activation | tanh |
Loss function | MSE |
Learning rate | \(1 \times 10^{-4}\) |
Max model parameters (\(n_t = 10\), \(m = 50\)) | 370,001 |
Min model parameters (\(n_t = 10\), \(m = 50\)) | 25,001 |