Statistics and Machine Learning to clarify the relationship (with supervised learning, for example) | entry 10mins | "statistical learning" study notes (a)

Statistical Learning

• Statistical Learning Features:

  Statistical learning (statistical learning) (statistical machine learning): Construction of computer data on a statistical model based on the law and the use of models to predict and analyze the data subject

  • Computer and network as a platform
  • The data for the study
  • The purpose is to predict and analyze data
  • To-centered approach, building models and applications
  • Probability theory, statistics, information theory, theory of computation, optimization theory and computer science and other areas of cross-disciplinary

• Statistical Learning Objects:

  • Data (similar data having a certain statistical regularity, such as random variables can be used to describe the characteristic data, statistical data described by the law of probability distribution)
  • In variable or set of presentation data. Represented by data types into continuous variables and discrete variables.

• Statistical learning process:

  • Extracting feature data
  • Abstract data model
  • Knowledge discovery in data
  • Data analysis and prediction

• Statistical Learning Objectives:

  • Forecast and analysis of data by constructing a probabilistic model

• Statistical learning methods:

  • Three elements: a hypothesis space model, model selection criteria and algorithm model of learning
    • Model (model), policy (strategy) and algorithms (algorithm)
  • classification:
    • Supervised learning (supervised learning): for classification, labeling and regression
      • From a given, limited, for learning the training data (training data) set start, assume that the data is generated iid
      • Determine the hypothesis space contains a collection of all possible models (hypothesis space), that is, learning model
      • Evaluation criteria to determine the model selected (evaluation criterion), that is, learning strategies
      • Achieve optimal model for solving the algorithm, learning algorithms
      • Selecting an optimum model from the hypothesis space, known training data and unknown test data (test data) have at best predict a given evaluation criteria
      • The optimal use of the learning model to predict or analyze new data
    • Unsupervised learning (unspervised learning)
    • Semi-supervised learning (semi-supervised learning)
    • Reinforcement learning (reinforcement learning)