27 notes in this chapter

Machine Learning and Deep Learning Terms MOC

27 core ML/DL concepts covering learning paradigms, training mechanics, neural network architectures, and evaluation. Each note has: definition, how it works, why it matters, common pitfalls, related terms, and a concrete example.

Core ML Concepts

• Supervised Learning
• Unsupervised Learning
• Reinforcement Learning
• Overfitting vs Underfitting
• Bias-Variance Tradeoff
• Cross-Validation
• Feature Engineering

Training Mechanics

• Gradient Descent
• Backpropagation
• Loss Function
• Learning Rate
• Epoch, Batch, and Iteration
• Regularization (L1, L2, Dropout)
• Hyperparameter Tuning

Neural Network Architectures

• Neural Network
• Convolutional Neural Network (CNN)
• Recurrent Neural Network (RNN)
• LSTM (Long Short-Term Memory)
• Activation Function
• Batch Normalization
• Vanishing-Exploding Gradient
• Transfer Learning
• Autoencoder
• GAN (Generative Adversarial Network)

Evaluation

• Confusion Matrix
• Precision, Recall, and F1 Score
• Ensemble Methods

---

How to use this

For the higher-level AI concepts these techniques power (LLMs, agents, RAG, alignment), see Artificial Intelligence MOC.

Suggested order if starting from zero

1. Supervised Learning → Unsupervised Learning → Reinforcement Learning — the three learning paradigms
2. Loss Function → Gradient Descent → Backpropagation → Learning Rate — how training actually works
3. Overfitting vs Underfitting → Bias-Variance Tradeoff → Regularization (L1, L2, Dropout) → Cross-Validation — generalization fundamentals
4. Neural Network → Activation Function → Convolutional Neural Network (CNN) → Recurrent Neural Network (RNN) → LSTM (Long Short-Term Memory) — architecture progression
5. Confusion Matrix → Precision, Recall, and F1 Score — evaluating classification models properly
6. Everything else (Transfer Learning, Autoencoder, GAN, Ensemble Methods) as you bump into it in projects or papers