Table 3.
Test set cross-modal loss with 95 % confidence intervals. TDImpute is compared to the DACMVA framework, namely the cross-modal variational auto-encoder (CM-VAE) and the CM-VAE with adversarial training (CM-VAE-Adv).
| Missing % | TDImpute | CM-VAE | CM-VAE-Adv |
|---|---|---|---|
| 4 | 0.0150 ± 0.0006 | 0.0144 ± 0.0007 | 0.0146 ± 0.0006 |
| 90 | 0.0165 ± 0.0007 | 0.0156 ± 0.0007 | 0.0158 ± 0.0007 |
| 95 | 0.0181 ± 0.0006 | 0.0164 ± 0.0007 | 0.0160 ± 0.0007 |