Geforce gtx 980 ti gpu

For comparison to MD-AD, we generate six analogous MLP networks with un-shared representations, and six linear models containing no hidden layers, to serve as baseline models (see Supplementary Figure 1a). All models were built using Tensorflow and Keras packages, and were constructed as consistently as possible, with the same inputs. The MLP baseline model was identical to the MD-AD model except with only a single branch of task-specific layers. Similarly, the linear baseline models were identical to the MLP baselines but with all hidden layers removed. Each model was trained on a single Nvidia GeForce GTX 980 Ti GPU. Although training time may vary across machines, we found that training the MD-AD model on the full dataset took about 350 seconds on average.

The machine learning methods used in this study, i.e., random forest,
decision tree, support vector machine, and kappa nearest neighbor,
are based on different concepts and are considered as state-of-art
for classification modeling.^{23 (link)} All the
BCMs and data analyses reported in this study were constructed using
KNIME (4.2.3) and Weka data mining software^{36 (link)} (cf. Supporting Information Figure S4
and Table S2). Plots were made using RStudio and Origin. A Microsoft
Windows 10, 64-bit based desktop computer, 8-core i7-6700K CPU@ 4.00GHz
with NVIDIA GeForce GTX980 Ti GPU was used.