Geforce rtx 2080 gpu

To avoid data-leakage when reporting the performance of the different models, nested stratified cross-validation was used. Nested cross-validation guarantees that different data is used to tune model parameters and to evaluate its performance by means of outer and inner cross-validation loops [47 ]. In the outer loop, train/test splits are generated, which are then used for averaging the test scores over several data splits. In the inner loop, the train set is further split in train/validation subsets. The best parameters are selected by minimizing the MEDAE on the validation splits. We used 5-folds and 7-folds stratified cross-validations in the outer and inner loops, respectively. To ensure that the distribution of RTs is representative of the population in all folds, stratification was performed by separating the target variable (RTs) into 6 different bins. The validation procedure is also summarized in Fig. S1 of Additional file 1.
The Bayesian hyperparameter search ("Bayesian hyperparameter search" Section) and the validation procedure described in this section approximately required 2.5 months of computational time in a computer with an AMD Ryzen Threadripper 2970WX with 24 cores at 1.85 Gz, and a NVIDIA GeForce RTX 2080 GPU.

All the computations reported in this study were performed on a Lenovo P620 workstation with AMD 3970X 32-Core processor, Nvidia GeForce RTX 2080 GPU, and 512GB of RAM.

We designed a convolutional neural network based on the well established U-Net architecture⁷⁶ for biomedical semantic image segmentation, as it recently proved its ability to perform well for 2D fetal brain MRI tissue segmentation^{22 (link)}. The baseline 2D U-Net is trained using a hybrid loss function defined as the sum of a categorical cross-entropy and a Dice loss. The latter is intended to mitigate any imbalance in the samples of the different classes^{22 (link),77} .
The implementation is performed in the framework of TensorFlow 2.5⁷⁸ and an Nvidia GeForce RTX 2080 GPU is deployed for training.

The server platform was configured as an Intel Core CPU I7-9700 K with a 3.60 GHz processor, 16 GB DDR4 2400Mhz memory, 2 TB hard drive capacity, 8 GB NVIDIA GeForce RTX 2080 GPU and the operating system was Windows 10.

We have trained our network solely on b0 images (15 per subject), using an 8-fold nested cross validation where we trained and validated on 27 subjects and tested on four. The proportion of the validation data was set to 15% of the training set. The training/validation set contains 25,920 slices of a 128 x 128 field of view, totaling 424,673,280 voxels. Our network was trained in an unsupervised manner by feeding normalized 2D axial slices that are encoded as feature maps in the latent space. The number of feature maps, and hence the dimensionality of the latent space, was optimized (optimal value to 32) using Keras-Tuner (52 ). The batch size and the learning rate were additionally optimized and set to 32 and 5e-5, respectively. The network that was initialized using (53 ) was trained for 200 epochs to minimize the mean squared error loss between the predicted and the ground truth image. We have utilized for this aim the Adam optimizer (54 (link)) with the default parameters β₁ = 0.5, β₂ = 0.999, and the network corresponding to the epoch with the minimal validation loss was then selected. The implementation was performed in the framework of TensorFlow 2.4.1 (55 ) and an Nvidia GeForce RTX 2080 GPU was deployed for training. Network code and checkpoint examples can be found in our Github repository¹.