Become an AI engineer
How models work, plainly
How models work, plainly is a unit of 29 lessons in Rabbithole's Become an AI engineer course, at levels 1 to 2.
Lessons
- Machine learning: computers that learn from dataLevel 1
- Neural networkLevel 1
- Deep learning, in plain wordsLevel 1
- Embeddings: turning words into numbersLevel 1
- Embeddings: giving words a place on a mapLevel 2
- Machine translation: from helper to translatorLevel 2
- When a model only memorisesLevel 2
- Supervised learning: learning from labeled examplesLevel 2
- Unsupervised learning: learning without labelsLevel 2
- Activation functions: the bend that lets networks learnLevel 2
- Chat templates: the layout a model expectsLevel 2
- Gradient descent: learning in small stepsLevel 2
- Reading Learning CurvesLevel 2
- Model Explainability: Why Did the Model Say That?Level 2
- Object Detection: Finding and Boxing Every ObjectLevel 2
- Reinforcement learning: learning by rewardLevel 2
- Softmax: from raw scores to chancesLevel 2
- Autoencoder: squeeze it, then rebuild itLevel 2
- Backpropagation: how a network finds what to changeLevel 2
- GANs: two AIs in a contestLevel 2
- Hyperparameters and the learning rateLevel 2
- Meta-Learning: Learning to LearnLevel 2
- Policy gradient: learning the rule itselfLevel 2
- Prediction-Based ExplorationLevel 2
- Recurrent neural networks: AI that reads one piece at a timeLevel 2
- Regularisation: keeping a model from memorisingLevel 2
- Residual connections: the shortcut around each stepLevel 2
- Q-Learning: A Table of Action ScoresLevel 2
- Deep Q-Learning: Swapping the Table for a NetworkLevel 2
Sources
These lessons draw on 25 sources. The most used: