NORDICS20 - Assets

Deep Learning on AWS

Amazon Web Services Resources EMEA

Issue link:

Contents of this Issue


Page 10 of 50

Amazon Web Services Deep Learning on AWS Page 6 Deep Learning Network Architecture • Multilayer Perceptrons (MLPs) (Feedforward neural networks [FFNNs]) • Convolutional Neural Networks (CNNs) • Recurrent Neural Networks (RNNs) At a high level, we chose a network architecture based on the specific use case that we are trying to solve. The following table is a decision matrix mapping use cases to the individual network architectures. Table 2: Mapping use cases to network architectures MLPs (FFNNs) CNNs RNNs (LSTM) Tabular Datasets Image Data Text Data Classification Prediction Problems Classification Prediction Problems Speech Data Regression Prediction Problems Regression Prediction Problems Classification Prediction Problems Regression Prediction Problems Deep Learning Algorithms Most deep learning models use gradient descent and backpropagation to optimize the neural network's parameters by taking partial derivatives of each parameter's contribution to the total change in error during the training process. Exploring optimization techniques concerning the training performance of deep learning algorithms is a topic of ongoing research and still evolving. Many new variants of the conventional gradient descent-based optimization algorithms such as momentum, AdaGrad (adaptive gradient algorithm), Adam (adaptive moment estimation), and Gadam (genetic-evolutionary Adam) have emerged to improve the learning performance of your deep learning network. Step 3. Set up and Manage the Environment for Training Designing and managing the deep learning environments for your training jobs can be challenging. Deep learning training jobs are different from traditional machine learning

Articles in this issue

view archives of NORDICS20 - Assets - Deep Learning on AWS