This post was published at 20 Apr 2019

Machine Learning

Machine Learning is a scientific discipline in the field of Artificial Intelligence that creates systems that learn automatically.
“Learning” in this context means identifying complex patterns in millions of data.
What the machine really learns is an algorithm that reviews the data and is able to predict or describe future behaviors.

Example:

It is said that a program can learn from an experience E with respect to some task T and performance P, if its performance in tasks T measured by P, improves the experience E.

  • Task: Is the case or problem to process.

  • Performance: Each P measured from task T.

  • Experience: Each result (improve or not) obtianed from a learning algrithm.

DS vs DA ds DM vs DAc

Difference between Data Science, Data Analysis, Data Mining and Data Analytics.

  • Data Science: Use Scientific methods, process and algorithms to extract knowlegde and insights from data (structured or/and unstructured). Data Science = Statistics + Data Analysis + Machine Learning + Data Mining.

  • Data Analysis: Is the process of inspecting, cleaning, transforming and modeling data with the goal of discovering useful information, conclusions and decision-making or relationship in the data.

  • Data Mining: Is the process of discovering patterns in large data sets for predictive purpose.

  • Data Analytics: Is the process of describe methods and logic for analysis.

Preprocessing process

  • Missing data: Set all NA values to the mean or just delete these examples.

  • Categorical data: When you have no comparable data, convert or encode labeling data in to numerical values. (useful in dependent variables).

    data Type A Type B Type C
    sample 1 1 0 0
    sample 2 0 0 1
    sample 3 0 1 0
  • Data set: Divide data into training-set, validation-set and test-set.

  • Feature Scale: Standarize(or Normalize) data (like distribution) to get a better organization of them.

Datasets

Datasets are the Open pre-organized data often used in machine learning, you can use it to train, evaluate and predict with your own data.

Machine Learning Classes

Is the different types of machine learning.

Info Supervised Learning UnSupervised Learning Reinforcement Learning
Data (X,Y) X state-actions
Goal Learn how to map x, y Learn how to identify structures and patterns to organize x Learn how to act in a certain environment based on rewards
Algorithms Regression/Classification Clustering/Dimension Reduction Model Free/Model Based
Example This is a cat This cat is like the tiger Cats are good to relax

Supervised Learning

This type is based on the term Conceptual Learning, in other words, the learning process is to find a function to map or relationate data X, Y.

See main article Supervised Learning.

Unsupervised Learning

This type is based on the term Self-Organized Learning, in other words, the learning process is to find a way to relationate patterns and characteristics of unorganized data X.

See main article UnSupervised Learning.

Reinforcement Learning

This type is based on the term Reinforcement Learning, in other words, the learning process is to maximize the reward obtained from every scenario modelated from data.

See main article Reinforcement Learning.

Machine Learning Types

Semi-Supervised Learning

This type is based on the term Inductive Learning, in other words, the learning process is to find a way to relationate a bit quantity of organized data to generalize it in large quantity of unorganized data.

See main article Semi-Supervised Learning.

Similarity Learning

This type is based on the term Compare Learning, in other words, the learning process is to find from previously organized data a similarity function that measures how similar or related two objects are.

See main article Similarity Learning.

Active Learning

This type is based on the term Query-Answer Learning, in other words, the learning process is the interaction between users and query website to get more data from them.

See main article Active Learning.

Transfer Learning

This type is based on the term Experience based Learning, in other words, the learning process is take a previous learned parameters from some pre-trained model and start new learning process using the target data.

See main article Transfer Learning.

Meta-Learning

This type is based on the term Autodidact Learning, in other words, the learning process is to understand how to “Learning to Learn” solving problems with greater flexibility, improving the performance of existing algorithms or inducing the learning algorithm itself.

See main article Meta-Learning.

Machine Learning Models/Algorithms

There is a different models or approach those define algorithms based on mathematics and statistics models in machine learning.

Neural Networks

Is the model based on human neurons to solve more complex problems.

See main article ANN.

Bayesian Networks

Is a model based on ANN with Bayesian statistics distributions.

See main article BNN.

Recursive Neural Networks

Is the model based on ANN and tree data structures to solve more complex problems.

See main article RcNN.

Recurrent Neural Networks

Is the model based on ANN to solve more complex problems.

See main article RNN.

Support Vector Machine

See main article SVM.

Complementary Information

Here you can find complementary information to better understand machine learning algorithms.

Activation Functions

Is the way to distribute data or scaling/normalize/standardize data to obtain a range of values to compare or classify.

See main article Activation Functions.

Regularization

Is the process to solve the overfitting problem.

See main article Regularization.