Table 7 Model performance and optimal hyperparameters under different latent space dimensions (Channels = 108).
From: Quality prediction using multiscale convolutional VAEs for thin plate parts
Dataset | Latent Dim | \(\:lr\) | \(\:{\lambda\:}_{KL}\) | \(\:{\lambda\:}_{1}\) | \(\:{\lambda\:}_{2}\) |
|---|---|---|---|---|---|
A | 8 | 0.02156 | 1.69e-02 | 1.69e-02 | 0.0666 |
16 | 0.0116 | 1.43e-02 | 1.43e-02 | 1.43e-02 | |
32 | 0.0256 | 3.09e-02 | 3.09e-02 | 0.0601 | |
64 | 0.0318 | 3.30e-02 | 3.30e-02 | 3.30e-02 | |
B | 8 | 0.0102 | 1.14e-02 | 1.14e-02 | 1.14e-02 |
16 | 0.0431 | 3.79e-02 | 3.79e-02 | 1.60e-01 | |
32 | 0.0587 | 5.25e-02 | 5.25e-02 | 5.25e-02 | |
64 | 0.0089 | 1.00e-02 | 1.00e-02 | 1.00e-02 | |
C | 8 | 1.03e-02 | 1.22e-02 | 1.22e-02 | 1.22e-02 |
16 | 9.30e-03 | 1.00e-02 | 1.00e-02 | 1.00e-02 | |
32 | 1.35e-02 | 1.00e-02 | 1.00e-02 | 1.00e-02 | |
64 | 3.44e-02 | 2.92e-02 | 2.92e-02 | 2.92e-02 |
