fast.ai: Practical Deep Learning for Coders

Jeremy Howard's top-down approach to deep learning. Build real models from day one.

About This Course

fast.ai takes a radically different approach to teaching deep learning. Instead of starting with theory, you build state-of-the-art models in the first lesson and gradually learn the underlying concepts. Jeremy Howard's philosophy is that the best way to learn is by doing.

The course uses the fastai library (built on PyTorch) and covers a remarkable breadth of deep learning applications. It is completely free and has produced many successful practitioners.

What You Will Learn

  • Computer Vision: Image classification, segmentation, object detection using CNNs
  • Natural Language Processing: Text classification, language models, sentiment analysis
  • Tabular Data: Neural nets for structured/tabular data, embeddings for categorical variables
  • Collaborative Filtering: Recommendation systems using deep learning
  • Training Techniques: Learning rate finder, one-cycle policy, mixed precision training
  • Data Augmentation: Transform pipelines, test-time augmentation
  • Model Deployment: Putting models into production, web apps, APIs
  • Ethics: Bias in ML, fairness, responsible AI development

Prerequisites

One year of coding experience (any language). No math prerequisites. The course teaches the math you need as you go.

Course content is freely available under the Apache 2.0 license.

Up next

Stanford CS231n: CNNs for Visual Recognition

Fei-Fei Li and Andrej Karpathy's deep learning course focused on convolutional neural networks and computer vision. Covers backprop, training tricks, detection, and segmentation.