Deep Learning in Practice
Neural networks, trained with discipline
Professor of Machine Learning · Dean of the Academy
About this course
Deep learning works when the practitioner respects the details: initialisation, normalisation, learning-rate schedules, data pipelines and the diagnostics that tell you which one is wrong. This course teaches those details through PyTorch, on datasets small enough to iterate on and large enough to matter.
What you will be able to do
- Build and train feed-forward and convolutional networks in PyTorch.
- Diagnose training pathologies from loss curves and gradient statistics.
- Apply transfer learning and fine-tuning responsibly.
Curriculum
8 lessons · 2h 39m
- 01
Networks from the ground up
- From logistic regression to a multilayer perceptronPreview20 min
- Backpropagation, explained once properly18 min
- Lab · Train an MLP on Fashion-MNIST40 min
- 02
Training discipline
- Initialisation, normalisation and schedules22 min
- Reading loss curves14 min
- Module quiz · Training8 min
- 03
Convolutions and transfer
- Convolutional networks21 min
- Transfer learning and fine-tuning16 min