Data, machine learning and deep learning
Deep learning with PyTorch course
Understand and build modern neural networks, from tensors to transformers.
Behind language models, computer vision and almost all modern AI are deep neural networks. PyTorch is the reference framework for building them, in research and in industry.
This course teaches deep learning by understanding what happens inside: tensors and autograd, the training loop, how to diagnose problems, convolutional networks and, as the highlight, the transformer architecture built from scratch up to a small language model of your own. Then we use pre-trained models to solve real problems with transfer learning and finish by deploying the model.
It is an intensive, hands-on course, with code in every session.
What you will learn
- Master the fundamentals of PyTorch and neural network training
- Build and train convolutional networks and transformers
- Apply transfer learning with pre-trained models
- Diagnose and improve training
- Export and deploy models
Syllabus
9 modules, from the fundamentals to the final project.
-
PyTorch fundamentals
- Tensors, operations and GPUs
- Automatic differentiation (autograd)
- Datasets and DataLoaders
-
Neural networks
- Multilayer perceptron and activation functions
- Loss functions and optimisers
- The training loop step by step
-
Training well
- Overfitting, regularisation and dropout
- Normalisation, learning rate and schedulers
- Experiment tracking
-
Vision with convolutional networks
- Convolutions and classic architectures (ResNet)
- Data augmentation
- Image classification
-
Sequences and attention
- Embeddings and recurrent networks
- The attention mechanism
-
Transformers
- The transformer architecture from scratch
- A small language model of your own
- Pre-trained models with Hugging Face
-
Transfer learning
- Fine-tuning vision and text models
- Freezing layers and adapting the head
-
Performance
- Mixed precision and torch.compile
- Multi-GPU training (introduction)
-
Deployment
- Saving and loading models
- Exporting to ONNX
- Serving a model through an API
Tools you will practise with
- Python
- PyTorch
- torchvision
- Hugging Face Transformers
- Jupyter or Google Colab
- ONNX
Hands-on project
You train and deploy a deep learning model on a real vision or text problem, comparing a model of your own with fine-tuning a pre-trained model, and publish it behind a simple API.
Requirements
- Fluent Python and machine learning knowledge (equivalent to the machine learning with Python course).
- Some linear algebra and calculus recommended.
- GPU access (cloud alternatives are provided).
Other courses
-
AI for data analysis course (no data science needed)
Draw conclusions from your data with AI, even if you are not a data scientist.
- Duration: 8 hours
- Level: Beginner
-
Computer vision with AI course
Teaching machines to see: inspection, counting, reading documents and more.
- Duration: 20 hours
- Level: Advanced
-
Demand forecasting and time series course
Anticipate demand, sales or consumption with models you can explain.
- Duration: 16 hours
- Level: Intermediate
Want this course for your team?
We adapt the content, length and format to your company.