BOOK Deep Learning with Python
Technology · Machine Learning
A practical course in deep learning with Keras, from tensors and training loops through vision, text, and generative models, including what current methods do not do.
Brett Gamble · professional · 60 hours · v0.1.3
About this course
This course follows the third edition of Deep Learning with Python by François Chollet and Matthew Watson. It is for someone who can already write Python and wants to train models, read their failures, and know when a neural network is the wrong tool.
The first seven modules build the foundations: what deep learning is, how tensors and gradient descent work, how TensorFlow, PyTorch, JAX, and Keras relate, and how to run classification, regression, evaluation, and the usual machine-learning workflow.
Later modules apply that workflow to images, timeseries, and text, then to generative models.
The last modules cover training and deployment practice, the limits of current methods, and a review.
Course learning objectives
- Explain what a deep learning model is doing when it maps data to a prediction, including the roles of tensors, loss, and gradient-based training.
- Frame a machine-learning task, choose an evaluation protocol, and tell underfitting from overfitting on a real dataset.
- Build, train, and evaluate a Keras model with the workflow that matches the problem, and read the training signals that decide the next change.
- Select and apply the architecture family the data calls for: convnets for images, sequence models for timeseries and text, and Transformers for language.
- Describe how a generative model produces new text or images, and what changes when the model is trained from scratch versus adapted from a pretrained model.
- State the practical and conceptual limits of these methods, including where scale does not replace adaptation to novelty.
Modules in this course
- What is deep learning? · 180 minutes
- The mathematical building blocks of neural networks · 220 minutes
- Introduction to TensorFlow, PyTorch, JAX, and Keras · 320 minutes
- Classification and regression · 276 minutes
- Fundamentals of machine learning · 195 minutes
- The universal workflow of machine learning · 255 minutes
- A deep dive on Keras · 307 minutes
- Image classification · 292 minutes
- ConvNet architecture patterns · 175 minutes
- Interpreting what ConvNets learn · 234 minutes
- Image segmentation · 184 minutes
- Object detection · 228 minutes
- Timeseries forecasting · 195 minutes
- Text classification · 180 minutes
- Language models and the Transformer · 180 minutes
- Text generation · 286 minutes
- Image generation · 330 minutes
- Best practices for the real world · 191 minutes
- The future of AI · 151 minutes
- Conclusions · 230 minutes
Request access
Access is approved by an administrator. Pricing is the same for every course — viewing this page does not enrol you.
Already approved? Sign in to enrol on the course page.