WIKIPEDIA

The free encyclopedia

Machine learning
Machine learning (ML); also called automated, computational, machine, or machine learning1 (from the
English machine learning, ML), is the subfield of computer
https://es.wikipedia.org/wiki/Ciencias_de_la_computaci%c3%b3nscience and a branch of artificial
intelligence, whose objective is to develop techniques that allow computers to learn. An agent is said to learn when
its performance improves with experience and through the use of data; that is, when the ability was not present in its
genotype or birth traits.2 "In machine learning, a computer observes data, builds a model based on that data, and
uses that model as both a hypothesis about the world and a piece of software that can solve problems." 3

In many cases, the field of action of machine learning overlaps with that of inferential statistics, since both
disciplines are based on data analysis. However, machine learning incorporates concerns about the computational
complexity of problems. Many problems are NP-hardhttps://es.wikipedia.org/wiki/NP-hard, so much of the
research in machine learning is focused on designing feasible solutions to these problems. Machine learning is also
closely related to pattern https://es.wikipedia.org/wiki/Reconocimiento_de_patronesrecognition. Machine
learning can be seen as an attempt to automate some parts of the scientific method using mathematical methods.
Therefore it is a process of knowledge induction.

Machine learning has a wide range of applications, including search engines, medical diagnostics, credit card fraud
detection, market analysis for different business sectors, DNA sequence classification, speech and written language
recognition, gaming and robotics.

New case
Summar
y
Some machine learning
systems attempt to remove
any need for intuition or
expert knowledge from data
Algorithm Model of:
analysis processes, while
Exampl Machine -Classification
others try to establish a learning -Optimization
e
collaborative framework
between the expert and the
computer. However, human
intuition cannot be
completely replaced, since
the designer of the system
has
Answer
to specify the form of data representation and the methods of data manipulation and characterization. However,
computers are used all over the world for very good technological purposes.

Models
Machine learning results in a model for solving a given task. Among the models, there are4 bookmark67

■ Geometric models, built in instance space and which can have one, two or multiple dimensions. If
there is a linear decision edge between the classes, the data is said to be linearly separable. A linear
decision boundary is defined as w * x = t, where w is a vector perpendicular to the decision boundary,
x is an arbitrary point on the decision boundary, and t is the decision threshold.
■ Probabilistic models, which attempt to determine the probability distribution describing the function
that links the values of the characteristics with given values. One of the key concepts for developing
probabilistic models is Bayesian statistics.
■ Logical models, which transform and express probabilities into rules organized in the form of decision
trees.
Models can also be classified as clustering models and gradient models. The former try to divide the instance space
into groups. The latter, as their name implies, represent a gradient in which one can differentiate between each
instance. Geometric classifiers such as support vector machines are gradient models.

Types of algorithms
See also: Adversarial machine learning
The different Machine Learning algorithms are grouped into a taxonomy
based on their output. Some types of algorithms are:

Supervised learning

The algorithm produces a function that establishes a

correspondence between the inputs and the desired outputs
of the system. An example of this type of algorithm is the
classificationhttps://es.wikipedia.org/wiki/Clasificaci
%c3%b3n_de_documentos problem, where the learning
system tries to label (classify) a series of vectors using one A support vector machine
of several categories (classes). The system's knowledge
base consists of examples of previous labeling. This type of
learning can be very useful in problems of biological research, computational biology and
bioinformatics.

Unsupervised learning

The entire modeling process is carried out on a set of examples consisting only of inputs to the
system. There is no information on the categories of these examples. Therefore, in this case, the
system has to be able to recognize patterns in order to label new entries.

Semi-supervised learning
This type of algorithm combines the two previous algorithms in order to classify appropriately. Both
marked and unmarked data are taken into account.

Reinforcement learning

The algorithm learns by observing the world around it. Your input is the feedback you get from the
 outside world in response to your actions. Therefore, the system learns by trial and error.

Reinforcement learning is the most general among the three categories. Instead of an instructor
telling the agent what to do, the intelligent agent must learn how the environment behaves through
rewards (reinforcements) or punishments, resulting from success or failure respectively. The main
objective is to learn the value function that helps the intelligent agent maximize the reward signal
and thus optimize its policies in order to understand the behavior of the environment and make
good decisions to achieve its formal objectives.

The main reinforcement learning algorithms are developed within the methods for solving finite
Markov decision https://es.wikipedia.org/wiki/Proceso_de_decisi%c3%b3n_de_M
%c3%a1rkovproblems, which incorporate Bellman equations and value functions. The three main
methods are: dynamic programming, Monte Carlo methods and temporal difference
https://es.wikipedia.org/wiki/Aprendizaje_por_diferencias_temporaleslearning.5

Among the implementations developed is AlphaGo, an AI program developed by Google

DeepMind to play the board game Go. In March 2016, AlphaGo won a match against professional
player Lee Se-Dol, who holds the rank of 9th dan and 18 world titles. Among the algorithms it uses
is the Monte Carlo search tree, it also uses deep learning with neural networks. You can see what
happened in the Netflix documentary “AlphaGo.”

Transduction
Similar to supervised learning, but does not explicitly build a function. It tries to predict the
categories of future examples based on the input examples, their respective categories, and the
examples new to the system.

Multi-task learning
Learning methods that use knowledge previously learned by the system to deal with problems
similar to those already seen.

The computational and performance analysis of machine learning algorithms is a branch of statistics known as
computational learning theory.

Machine learning is carried out automatically by humans because it is such a simple process for us that we do not
even realize how it is done and everything it involves. From the moment we are born until we die, human beings
carry out different processes, among them we find the learning process through which we acquire knowledge,
develop skills to analyze and evaluate through methods and techniques as well as through our own experience.

However, machines must be taught how to learn, because if a machine cannot develop its skills, the learning process
will not take place, but will only be a repetitive sequence.

Classification techniques

Decision trees
This type of learning uses a decision tree as a predictive model. Observations about an object are mapped to
conclusions about the final value of that object.

Trees are basic structures in computer science. Attribute trees are the basis for decisions. One of the two main forms
of decision trees is the one developed by Quinlan of measuring the impurity of
entropyhttps://es.wikipedia.org/wiki/Entrop%c3%ada_(informaci%c3%b3n) in each branch, something he
first developed in the ID3 algorithm and then in C4.5https://es.wikipedia.org/wiki/C4.5. Another strategy is
based on the GINI index. The CART algorithm is an implementation of this strategy. 6

Association rules
Association rule algorithms seek to discover interesting relationships between variables. Among the best-known
methods are the a priori algorithm, the Eclat algorithm, and the frequent pattern algorithm.

Genetic algorithms
Genetic algorithms are heuristic search processes that simulate natural selection. They use methods such as
mutation and crossover to generate new classes that can offer a good solution to a given problem.

Artificial neural networks

Artificial neural networks (ANNs) are a machine learning paradigm inspired by neurons in animal nervous systems.
It is a system of neuron links that collaborate with each other to produce an output stimulus. Connections have
numerical weights that are adapted based on experience. In this way, neural networks adapt to an impulse and are
able to learn. The importance of neural networks fell for a while with the development of support vectors and linear
classifiers, but re-emerged in the late 2000s with the advent of deep learning.

Support Vector Machines

SVMs are a set of supervised learning methods used for classification and regression. MVS algorithms use a
training set of examples classified into two categories to build a model that predicts whether a new example belongs
to one or the other of those categories.

Clustering algorithms

Clustering analysis is the classification of observations into subgroups—clusters—so that the observations in each
group resemble each other according to certain criteria.

Clustering techniques make different inferences about the structure of the data; they are usually guided by a specific
similarity measure and by a level of internal compactness (similarity between members of a group) and separation
between different groups.

Clustering is an unsupervised learning method and is a very popular technique for statistical data analysis.

Bayesian networks
A Bayesian network, belief network or directed acyclic model is a probabilistic model that represents a series of
random variables and their conditional independence through a directed acyclic graph. A Bayesian network can
represent, for example, the probabilistic relationships between diseases and symptoms. Given certain symptoms, the
network can be used to calculate the probabilities that certain diseases are present in an organism. There are
efficient algorithms that infer and learn using this type of representation.
Knowledge
In machine learning we can obtain 3 types of knowledge, which are:

1. Growth
It is what is acquired from what surrounds us, which stores information in memory as if it left
traces.
2. Restructuring
When interpreting knowledge, the individual reasons and generates new knowledge, which is
called restructuring.
3. Adjustment
It is obtained by generalizing several concepts or generating your own.

All three types are carried out during a machine learning process but the importance of each type of knowledge
depends on the characteristics of what is being learned.

Learning is more than a necessity, it is a key factor in meeting the needs of artificial intelligence.

Distinction between supervised and unsupervised learning

Supervised learning is characterized by having information that specifies which data sets are satisfactory for the
learning objective. An example could be software that recognizes whether or not a given image is an image of a
face: to learn the program we would have to provide it with different images, specifying in the process whether or
not they are faces.

In unsupervised learning, however, the program does not have data that defines what information is satisfactory or
not. The main objective of these programs is usually to find patterns that allow data to be separated and classified
into different groups, based on their attributes. Following the previous example, unsupervised learning software
would not be able to tell us whether a given image is a face or not, but it could, for example, classify images
between those that contain human faces, those that contain animals, or those that do not contain faces. The
information obtained by an unsupervised learning algorithm must subsequently be interpreted by a person to make
it useful.

Applications
■ Search Engines
■ Medical diagnosis
■ Detecting fraud using credit cards
■ Stock Market Analysis
■ Classification of DNA sequences
■ Feature engineering
■ Speech recognition
■ Robotics
■ Data mining
■ Big Data
■ Time series forecasts
Machine Learning Topics
Below are a number of topics that could be part of the syllabus of a machine learning course.

■ Modeling conditional probability density functions: classification and regression

■ Artificial neural networks
■ Decision Trees: Decision tree learning uses a decision tree as a predictive model that maps
observations to conclusions about the value of a given object.
■ Unpostulated multiple regression models
■ Regression in Gaussian processes
■ Linear discriminant analysis
■ k-nearest neighbors
■ Perceptron
■ Radial basis functions
■ Support vector machines
■ Modeling probability density functions using generative models
■ EM algorithm
■ Graphical models, such as Bayesian https://es.wikipedia.org/wiki/Red_bayesiananetworks and
Markov random https://es.wikipedia.org/wiki/Campo_aleatorio_de_Markovfields
■ Generative Topographic Mapping
 ■ Approximate inference techniques
■ Markov Chains and Monte Carlo Method
■ Variational methods
■ Optimization: Most of the methods described above use optimization algorithms or are themselves
instances of optimization problems.

History and relationship with other topics

Machine learning was born from the quest for artificial intelligence. Already in the early days of AI as an academic
discipline, some researchers were interested in making machines learn. They tried to solve the problem with various
symbolic methods, as well as what they called 'neural networks' which were generally perceptrons and other models
basically based on generalized linear models as known in statistics.

Software
Many programming languages can be used to implement machine learning algorithms. The most popular in 2015
were R and Python.7 R is widely used primarily in academia, while Python is more popular in the private sector.

Software packages that include machine learning algorithms include:

Open source software

■ TensorFlow: platform ■ ELKI: a platform for Java related to
multilingual licensed under AGPLv3 the
■ Shrink machine learning
and (e1071, rpart, nnet,
■ H2O
multiplatform randomForest, among
■ KNIME others)
developed by Google and
licensed as Apache 2. ■ mlpy ■ RapidMiner
■ Apache Mahout: ■ MLPACK ■ scikit-learn: Python library
Java platform for scalable ■ MOA that interacts with NumPy
machine learning algorithms, ■ OpenCV and SciPy
especially in the areas of ■ JS Omelette ■ Spark MLlib: a library that is
collaborative filtering, ■ OpenNN part of Apache Spark, a
clustering, and classification platform for
■ A: language of
■ dlib: a Boost-licensed library Statistical programming with cluster computing
for developing in C++ numerous libraries ■ Weka: a library in Java

Commercial software
■ SPSS Modeler ■ Microsoft Azure - Oracle Data Mining
■ Mathematica Machine Learning - RCASE
■ MATLAB ■ Neural Designer - STATISTICS
■ SAS
Biases
Machine learning algorithms can often be affected by bias in the data (See algorithmic bias). For example, not all
entries for which no information was received in the training phase can be classified. In fact, when training is done
with data classified by humans, machine learning tends to create the same biases that exist in society. Some
examples of this are when in 2015 the Google Photos algorithm identified some black people as gorillas, or in 2016
when Microsoft's Twitter bot developed racist and sexist behavior based on observing data traffic on said social
network. For this reason, in recent years there has been a trend towards developing methods to increase
fairnesshttps://es.wikipedia.org/wiki/Equidad_(aprendizaje_autom%c3%a1tico), that is, to reduce bias in this
type of algorithms by AI experts. Quoting Fei-fei Li "There is nothing special about AI. It is inspired by people,
created by people, and most importantly impacts people. It is a very powerful tool that we have only just begun to
understand, and that is a great responsibility" 8

See also
■ Adversarial Machine Learning
■ Deep learning
■ Systems dynamics
■ Artificial intelligence
■ Computational Intelligence
■ Internet of Things
■ Dynamic system
■ Pattern recognition
■ Association rules
■ Autonomous robot
■ Fairness (machine learning)
■ Ablation (artificial intelligence)
■ OpenAI Codex
■ Fawkes (image cloaking software)
■ Residual neural network
■ Hyperparameter (machine learning)
■ Learning by temporal differences
■ Ensemble learning
■ Rule-based machine learning
■ Statistical learning theory
■ Bias-variance dilemma
■ Ockham learning

References
1. «Brief introduction to the AI world. Chapter 1/10» (https://www.linkedin.com/pulse/breve-intro ducci
%C3%B3n-al-mundo-ia-ram%C3%B3n-garc%C3%ADa-espeleta-rgu2f).
www.linkedin.com. Accessed May 25, 2024.
2. Russell, Stuart; Norvig, Peter (2009). Artificial Intelligence: A Modern Approach (3rd edition). p. 229.
3. Russell and Norvig (2021). Artificial Intelligence: A Modern Approach. Pearson. p. 651. ISBN
9780134610993.
4. Flach 2012 pp. 20-21
5. Sutton, Richard S., Barto, Andrew G. Reinforcement Learning: An Introduction. The MIT Press.
6. Flach 2012 pp. 155-156
7. Four main languages for analytics and data mining science (KD Nuggets) (http://www.kdnug
gets.com/2014/08/four-main-languages-analytics-data-mining-data-science.html)
8. «Fei-Fei Li's Quest to Make Machines Better for Humanity» (https://www.wired.com/story/fei-f ei-li-
artificial-intelligence-humanity/). Wired (in English). ISSN 1059-1028 (https://portal.issn.org/resour
ce/issn/1059-1028). Accessed on December 17, 2019.

Literature
■ Bishop, Christopher (2008) Pattern Recognition and Machine Learning. Springer Verlag. ISBN=978-
0-3873-1073-2.
■ Flach, Peter (2012) Machine Learning: The Art and Science of Algorithms that Make Sense of Data.
Cambridge University Press. ISBN 978-1-107-42222-3.
■ Gollapudi, Sunila (2016) Practical Machine Learning. Packt Publishing. ISBN=978-1-78439- 968-4.
■ Ian H. Witten and Eibe Frank (2011). Data Mining: Practical machine learning tools and techniques
Morgan Kaufmann, 664 pages, ISBN 978-0-12-374856-0.
■ Mitchell, T. (1997). Machine Learning, McGraw Hill. ISBN 0-07-042807-7
■ Raschka, Sebastian (2015). Python Machine Learning, Packt Open Source. ISBN 978-178355-513-0

External links
■ Practical examples of Machine Learning in Spanish (https://machinelearningenespanol.com/blog/)
■ Machine Learning Blog - The Machine Learning Bible (http://www.rubenjromo. com/)
■ Machine Learning will change the world (https://www.emerita.legal/blog/machine-learning-will-
change-the-world/)
■ Machine Learning Development with Perl (http://www.perlmonks.org/index.pl?node_id=6383 91) (in
English)
■ Study and application of machine learning techniques oriented to the medical field: estimation and
explanation of individual predictions. Autonomous University of Madrid (htt
ps://repositorio.uam.es/bitstream/handle/10486/12100/59264_Di_Deco_Sampedro_JavierP FM.pdf?
sequence=1)
■ AlphaGo (https://www.netflix.com/title/80190844) Archived (https://web.archive.org/web/201
80204185620/https://www.netflix.com/title/80190844) February 4, 2018, at Wayback Machine.
■ Machine Learning explained (podcast) (https://pochocosta.com/podcast/9-machine-learning-
explained/)https://pochocosta.com/podcast/9-machine-learning-explicado/
■ Machine Learning: Selecting Classification Metrics (in Spanish)
(http://sitiobigdata.com/2019/01/19/machine-learning-metrica-clasificacion-parte-3/)

Obtained from «https://en.wikipedia.org/w/index.php?title=Machine_learning&oldid=160992708»