A path from intuition
to engineering.

From Machine Learning to Deep Learning

What changes when models learn representations from raw data?

• 1.1 Where Machine Learning Runs Out of Road102 min
• 1.2 The Learning Mechanism, Traced By Hand128 min
• 1.3 Your First PyTorch Network121 min

PyTorch: Building Neural Networks

How do I actually build a deep learning model?

• 2.1 Tensors and the Shape Contract149 min
• 2.2 Dataset and DataLoader: Feeding the Model100 min
• 2.3 Building and Training an MLP You Can Defend117 min
• 2.4 From Notebook to Project76 min

Training Dynamics & Optimization

Why does a network learn well, badly, or not at all?

• 3.1 Optimizers and the Learning Rate83 min
• 3.2 Overfitting, and What to Do About It80 min
• 3.3 When Gradients Misbehave65 min
• 3.4 The Diagnosis Clinic80 min

Deep Learning Engineering

How do I run deep learning experiments professionally?

• 4.1 Data You Can Trust108 min
• 4.2 Experiments You Can Reproduce74 min
• 4.3 Memory, Batch Size and Precision79 min
• 4.4 The Bake-Off: Gradient Boosting against a Neural Network71 min

Learning Representations from Images and Text

How do neural networks learn useful representations from different types of data?

• 5.1 Understanding Representation LearningClassroom · 35 min
• 5.2 How Neural Networks Learn from ImagesClassroom · 61 min
• 5.3 How Neural Networks Learn from TextClassroom · 51 min
• 5.4 From Sequences to AttentionClassroom · 41 min
• 5.5 Connecting Image and Text RepresentationsClassroom · 32 min
• 5.6 Synthesis: Putting Representation Learning into PracticeClassroom · 26 min

Transformers and Attention Mechanisms

How does a Transformer process information and learn relationships between elements in a sequence?

Lessons 6.1–6.6, including architecture history, vision and multimodal applications, and weight tying, are covered in class.

• 6.1 Understanding the Transformer ArchitectureClassroom · 34 min
• 6.2 Preparing Inputs for a TransformerClassroom · 27 min
• 6.3 Understanding AttentionClassroom · 46 min
• 6.4 The Transformer BlockClassroom · 33 min
• 6.5 Building a Decoder-Only TransformerClassroom · 24 min
• 6.6 How Transformers Learn and Generate TextClassroom · 47 min

Pretrained Models and the Hugging Face Ecosystem

How can AI engineers leverage models that have already learned useful representations?

Lessons 7.1–7.5 are classroom material. Lessons 7.6–7.7 are complete further-learning notes.

• 7.1 Understanding Pretrained ModelsClassroom · 56 min
• 7.2 The Hugging Face Ecosystem (Essentials)Classroom · 47 min
• 7.3 Finding and Selecting the Right ModelClassroom · 39 min
• 7.4 Loading and Using Pretrained ModelsClassroom · 43 min
• 7.5 Building Applications with Pretrained ModelsClassroom · 48 min
• 7.6 Inference and Deployment ConstraintsFurther learning · 48 min
• 7.7 Practical Model Selection and IntegrationFurther learning · 36 min

Efficient Fine-Tuning and Data Engineering

How can we adapt pretrained models efficiently and evaluate whether adaptation is worthwhile?

Lessons 8.1–8.5 and 8.7 are classroom material. Lesson 8.6 extends the workflow to LLM instruction tuning and QLoRA; section 7 of 8.5 is an extension on other PEFT methods.

• 8.1 Understanding Model AdaptationClassroom · 31 min
• 8.2 Preparing Data for Fine-TuningClassroom · 44 min
• 8.3 Establishing BaselinesClassroom · 25 min
• 8.4 The Fine-Tuning WorkflowClassroom · 40 min
• 8.5 Parameter-Efficient Fine-Tuning (PEFT)Classroom · 43 min
• 8.6 Fine-Tuning Large Language Models (Conceptual)Further learning · 37 min
• 8.7 Evaluating Adapted ModelsClassroom · 37 min