A model is a learned approximation. It takes input, applies a mathematical transformation, and produces output. The transformation is shaped by training, and its quality is judged by how well it generalizes to new data.

Models come in many forms. Linear regression fits a line. Decision trees split data recursively. Neural networks compose layers of learned functions. Each has strengths, weaknesses, and assumptions about the data.

What defines a model

• Architecture and parameter count
• Training procedure and objective
• Input and output format
• Performance on held-out data
• Computational and memory requirements

No model is correct in an absolute sense. A model is useful or not for a specific task, dataset, and deployment context. The same architecture can succeed on one problem and fail on another.