Andromeda
Note

Machine Learning

Definition

Machine Learning (ML) is a subfield of artificial intelligence focused on algorithms and statistical models that enable computers to perform tasks by optimizing an objective (loss) function over experience (data and feedback) rather than relying on explicit per-case programming.

Why It Matters

Machine learning is the general-purpose technology of the 21st century. Understanding loss landscapes, gradients, generalization, and dimensionality is prerequisite to scaling systems, evaluating capability/safety claims, and participating in the governance of transformative AI rather than merely being affected by it.

Core Concepts

  • Objective / Loss Function: A scalar measure J(heta)J( heta) of how bad the model’s predictions are on the training data. Training optimizes this landscape.
  • Gradient Descent Parameter Update: The fundamental local optimization step: θ:=θ−α∇θJ(θ)\theta := \theta - \alpha \nabla_\theta J(\theta) How to read: “Theta is updated to theta minus alpha, the learning rate, times the gradient of J with respect to theta.” Meaning / when to use: Move the parameters in the direction that most steeply decreases the loss, scaled by step size alpha. The workhorse of almost all modern ML training.
  • Generalization Gap: The difference between performance on training data and performance on new, unseen data. The central problem of ML.
  • Overfitting & Underfitting: Overfitting is memorizing noise (high variance); Underfitting is failing to capture signal (high bias).
  • Curse of Dimensionality: As input dimension grows, data becomes sparse, which techniques like PCA, SVD, and deep representation learning combat by finding lower-dimensional structures.
  • Paradigms: Supervised learning (labeled data), Unsupervised learning (finding unlabeled structure), and Reinforcement learning (reward signals).

Connected Concepts

Connected notes