DataRoot University by DRL
Practical Deep Learning with PyTorch
Build and train neural networks with PyTorch. Explore computer vision, sequence models, generative models and transformers through practical assignments, then learn how to deploy your models.
- 16 weeks
- Online · Slack
- Free
Before you start
Confident Python and basic machine learning. Data Science Fundamentals covers the preparation needed for this course.
What you’ll build
An image segmentation pipeline and a face-editing application using StyleGAN, alongside assignments in sequence modelling and generation.
Course outline
8 modulesNeural network foundations
Connect the mathematics of neural networks to their implementation. Study hyperparameter tuning and regularisation, then complete a module test.
Working with PyTorch
Build, train and evaluate networks. Explore automatic differentiation and practise the training pipeline in two labs.
Convolutional neural networks
Tackle image classification, detection and segmentation with architectures including ResNet, YOLOv3 and U-Net. Apply them in labs and a Kaggle competition.
Recurrent neural networks
Work with RNNs, GRUs and LSTMs for sequential data. Generate music with a character-level model and explore human activity recognition.
Autoencoders
Study representation learning, compare autoencoders with PCA, and train a denoising autoencoder on a real dataset.
Generative models
Explore GANs, Wasserstein GANs, progressive training and StyleGAN. Practise training and evaluating generators, including a CryptoPunks assignment.
Attention & transformers
Study sequence-to-sequence models and attention. Explore transformer applications with BERT and GPT-2, then complete a practical project.
Deploying PyTorch models
Prepare trained networks for use outside the training environment. Explore model serialisation and optimisation for inference.