Table 9
Comparative analysis of time and memory across models.
| Model type | Architecture | Bottleneck | GPU (ms) | CPU (ms) | Params (M) | |
|---|---|---|---|---|---|---|
| Legend descriptor | ||||||
| β-VAE (U*) (1) | β-VAE | CNN–CNN | 128 | 0.386 | 113.8 | 7.0 |
| infoVAE (mix) | infoVAE | CNN-ResNet | 32 | 3.970 | 779.6 | 14.0 |
| AE (mix) | AE | CNN-ResNet | 9 | 3.910 | 795.0 | 13.3 |
| CNN (mix) | encoder | CNN | 6 | 0.190 | 38.4 | 1.5 |
| CNN (simulation) | decoder | CNN | 6 | 0.217 | 76.3 | 1.7 |
| ResNet (simulation) | decoder | ResNet | 6 | 3.694 | 729.9 | 11.6 |
Notes. The “GPU” column shows the time required to process a single spectrum using a GPU, measured as the time to process a batch divided by the batch size. Similarly, the “CPU” column shows the results of the same experiment but using a CPU. The “params” column represents the number of the ML parameters, expressed in millions. (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.