Table 6
Utilizing outputs from unsupervised models for label prediction.
| (a) Model descriptions | ||||
|---|---|---|---|---|
| Learning paradigm | Architecture | Bottleneck | Dataset | |
| Legend descriptor | ||||
| betaVAE (U*) (1) | Unsupervised | CNN-CNN | 128 | Real |
| betaVAE (U) | Unsupervised | CNN-ResNet | 128 | Real |
| infoVAE (U) | Unsupervised | CNN-ResNet | 128 | Real |
| betaVAE (mix, U) | Unsupervised | CNN-ResNet | 128 | Mix |
| infoVAE (mix, U) | Unsupervised | CNN-ResNet | 128 | Mix |
| (b) Intrinsic labels | ||||
| Teff (K) | [M/H] (dex) | log(g) (dex) | All (−) | |
| betaVAE (U*) (1) | 978.0 ± 13.0 | 0.1578 ± 0.0027 | 0.1798 ± 0.0025 | 1.548 ± 0.013 |
| betaVAE (U) | 986.0 ± 13.0 | 0.1579 ± 0.0027 | 0.1829 ± 0.0025 | 1.563 ± 0.013 |
| infoVAE (U) | 999.0 ± 13.0 | 0.1529 ± 0.0027 | 0.1817 ± 0.0025 | 1.562 ± 0.013 |
| betaVAE (mix, U) | 1017.0 ± 13.0 | 0.154 ± 0.0027 | 0.186 ± 0.0025 | 1.59 ± 0.013 |
| infoVAE (mix, U) | 990.0 ± 13.0 | 0.1648 ± 0.0026 | 0.1846 ± 0.0025 | 1.583 ± 0.013 |
| (c) Extrinsic labels | ||||
| Radvel (km s−1) | BERV (km s−1) | Airmass (−) | All (−) | |
| betaVAE (U*) (1) | 31.07 ± 0.27 | 16.01 ± 0.096 | 0.181 ± 0.0019 | 1.5751 ± 0.0074 |
| betaVAE (U) | 32.15 ± 0.28 | 16.125 ± 0.097 | 0.1827 ± 0.0019 | 1.6056 ± 0.0075 |
| infoVAE (U) | 31.26 ± 0.28 | 16.077 ± 0.097 | 0.181 ± 0.0019 | 1.5811 ± 0.0075 |
| betaVAE (mix, U) | 31.72 ± 0.28 | 16.265 ± 0.1 | 0.1798 ± 0.0019 | 1.5935 ± 0.0076 |
| infoVAE (mix, U) | 31.81 ± 0.28 | 16.39 ± 0.1 | 0.1809 ± 0.0019 | 1.6017 ± 0.0076 |
Notes. The column titled “All” uses NMAE from Eq. (13) to summarize the mean absolute error across all normalized labels. The notation “a ± b” represents the mean absolute error ± the standard deviation for each respective label and model. (1) Sedaghat et al. (2021).
Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.
Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.
Initial download of the metrics may take a while.