Statistical LearningBuilding probabilistic statistical models based on data and using them to predict and analyze data is also known as statistical machine learning.

The premise of statistical learning is that the basic assumption of the data is that the data is of the same type and has certain statistical regularities; based on the data, starting from the data, extract the characteristics of the data, abstract the data model, discover the laws therein, find predictive functions and solve the problem; its purpose is to consider what kind of model to learn and how to learn the model.

Statistical learning is a data-driven discipline and an interdisciplinary subject that integrates multiple fields such as probability theory, statistics, information theory, computational theory, optimization theory, and computer science.

Three Elements of Statistical Learning

• Model: The choice of generative model or discriminative model. The difference between these two models is that the goal is different. The generative model is to find the joint probability distribution of the source data, while the discriminative model is to find the conditional probability or decision function;
• Strategy: Choose a suitable loss function or risk function, that is, choose an objective function;
• Algorithms: including gradient descent method, Newton method/quasi-Newton method, Lagrangian method and other optimization algorithms.