text
stringlengths
2
350
source
stringlengths
9
49
* [Friendly AI](https://en.wikipedia.org/wiki/Friendly_artificial_intelligence "Friendly artificial intelligence") * [Control problem](https://en.wikipedia.org/wiki/AI_control_problem "AI control problem")/[Takeover](https://en.wikipedia.org/wiki/AI_takeover "AI takeover")
ml.md_0_164
* [Ethics](https://en.wikipedia.org/wiki/Ethics_of_artificial_intelligence "Ethics of artificial intelligence") * [Existential risk](https://en.wikipedia.org/wiki/Existential_risk_from_artificial_general_intelligence "Existential risk from artificial general intelligence")
ml.md_0_165
* [Turing test](https://en.wikipedia.org/wiki/Turing_test "Turing test") * [Uncanny valley](https://en.wikipedia.org/wiki/Uncanny_valley "Uncanny valley")
ml.md_0_166
[History](https://en.wikipedia.org/wiki/History_of_artificial_intelligence "History of artificial intelligence") * [Timeline](https://en.wikipedia.org/wiki/Timeline_of_artificial_intelligence "Timeline of artificial intelligence")
ml.md_0_167
* [Progress](https://en.wikipedia.org/wiki/Progress_in_artificial_intelligence "Progress in artificial intelligence") * [AI winter](https://en.wikipedia.org/wiki/AI_winter "AI winter") * [AI boom](https://en.wikipedia.org/wiki/AI_boom "AI boom")
ml.md_0_168
Glossary * [Glossary](https://en.wikipedia.org/wiki/Glossary_of_artificial_intelligence "Glossary of artificial intelligence")
ml.md_0_169
* [v](https://en.wikipedia.org/wiki/Template:Artificial_intelligence "Template:Artificial intelligence") * [t](https://en.wikipedia.org/wiki/Template_talk:Artificial_intelligence "Template talk:Artificial intelligence")
ml.md_0_170
* [e](https://en.wikipedia.org/wiki/Special:EditPage/Template:Artificial_intelligence "Special:EditPage/Template:Artificial intelligence")
ml.md_0_171
**Machine learning** (**ML**) is a [field of study](https://en.wikipedia.org/wiki/Field_of_study "Field of study") in [artificial intelligence](https://en.wikipedia.org/wiki/Artificial_intelligence "Artificial intelligence") concerned with the development and study of [statistical algorithms](https://en.wikipedia.org/wiki/Computational_statistics
ml.md_0_172
"Computational statistics") that can learn from [data](https://en.wikipedia.org/wiki/Data "Data") and [generalise](https://en.wikipedia.org/wiki/Generalise "Generalise") to unseen data, and thus perform [tasks](https://en.wikipedia.org/wiki/Task_\(computing\) "Task \(computing\)") without explicit
ml.md_0_173
\(computing\)") without explicit [instructions](https://en.wikipedia.org/wiki/Machine_code "Machine code").[[1]](https://en.wikipedia.org/wiki/Machine_learning#cite_note-1) Within a subdiscipline in machine learning, advances in the field of [deep learning](https://en.wikipedia.org/wiki/Deep_learning "Deep learning") have allowed [neural
ml.md_0_174
learning") have allowed [neural networks](https://en.wikipedia.org/wiki/Neural_network_\(machine_learning\) "Neural network \(machine learning\)"), a class of statistical algorithms, to surpass many previous machine learning approaches in performance.[[2]](https://en.wikipedia.org/wiki/Machine_learning#cite_note-ibm-2)
ml.md_0_175
ML finds application in many fields, including [natural language processing](https://en.wikipedia.org/wiki/Natural_language_processing "Natural language processing"), [computer vision](https://en.wikipedia.org/wiki/Computer_vision "Computer vision"), [speech recognition](https://en.wikipedia.org/wiki/Speech_recognition "Speech recognition"),
ml.md_0_176
"Speech recognition"), [email filtering](https://en.wikipedia.org/wiki/Email_filtering "Email filtering"), [agriculture](https://en.wikipedia.org/wiki/Agriculture "Agriculture"), and [medicine](https://en.wikipedia.org/wiki/Medicine
ml.md_0_177
"Medicine").[[3]](https://en.wikipedia.org/wiki/Machine_learning#cite_note-tvt-3)[[4]](https://en.wikipedia.org/wiki/Machine_learning#cite_note-YoosefzadehNajafabadi-2021-4) The application of ML to business problems is known as [predictive analytics](https://en.wikipedia.org/wiki/Predictive_analytics "Predictive analytics").
ml.md_0_178
[Statistics](https://en.wikipedia.org/wiki/Statistics "Statistics") and [mathematical optimisation](https://en.wikipedia.org/wiki/Mathematical_optimisation "Mathematical optimisation") (mathematical programming) methods comprise the foundations of machine learning. [Data mining](https://en.wikipedia.org/wiki/Data_mining "Data mining") is a related
ml.md_0_179
"Data mining") is a related field of study, focusing on [exploratory data analysis](https://en.wikipedia.org/wiki/Exploratory_data_analysis "Exploratory data analysis") (EDA) via [unsupervised learning](https://en.wikipedia.org/wiki/Unsupervised_learning "Unsupervised
ml.md_0_180
"Unsupervised learning").[[6]](https://en.wikipedia.org/wiki/Machine_learning#cite_note-6)[[7]](https://en.wikipedia.org/wiki/Machine_learning#cite_note-Friedman-1998-7)
ml.md_0_181
From a theoretical viewpoint, [probably approximately correct learning](https://en.wikipedia.org/wiki/Probably_approximately_correct_learning "Probably approximately correct learning") provides a framework for describing machine learning. ## History
ml.md_0_182
## History [[edit](https://en.wikipedia.org/w/index.php?title=Machine_learning&action=edit&section=1 "Edit section: History")] See also: [Timeline of machine learning](https://en.wikipedia.org/wiki/Timeline_of_machine_learning "Timeline of machine learning")
ml.md_0_183
The term _machine learning_ was coined in 1959 by [Arthur Samuel](https://en.wikipedia.org/wiki/Arthur_Samuel_\(computer_scientist\) "Arthur Samuel \(computer scientist\)"), an [IBM](https://en.wikipedia.org/wiki/IBM "IBM") employee and pioneer in the field of [computer gaming](https://en.wikipedia.org/wiki/Computer_gaming "Computer gaming") and
ml.md_0_184
"Computer gaming") and [artificial intelligence](https://en.wikipedia.org/wiki/Artificial_intelligence "Artificial intelligence").[[8]](https://en.wikipedia.org/wiki/Machine_learning#cite_note-Samuel-8)[[9]](https://en.wikipedia.org/wiki/Machine_learning#cite_note-Kohavi-9) The synonym _self-teaching computers_ was also used in this time
ml.md_0_185
was also used in this time period.[[10]](https://en.wikipedia.org/wiki/Machine_learning#cite_note-cyberthreat-10)[[11]](https://en.wikipedia.org/wiki/Machine_learning#cite_note-11)
ml.md_0_186
Although the earliest machine learning model was introduced in the 1950s when [Arthur Samuel](https://en.wikipedia.org/wiki/Arthur_Samuel_\(computer_scientist\) "Arthur Samuel \(computer scientist\)") invented a [program](https://en.wikipedia.org/wiki/Computer_program "Computer program") that calculated the winning chance in checkers for each
ml.md_0_187
chance in checkers for each side, the history of machine learning roots back to decades of human desire and effort to study human cognitive processes.[[12]](https://en.wikipedia.org/wiki/Machine_learning#cite_note-WhatIs-12) In 1949, [Canadian](https://en.wikipedia.org/wiki/Canadians "Canadians") psychologist [Donald
ml.md_0_188
"Canadians") psychologist [Donald Hebb](https://en.wikipedia.org/wiki/Donald_O._Hebb "Donald O. Hebb") published the book _[The Organization of Behavior](https://en.wikipedia.org/wiki/Organization_of_Behavior "Organization of Behavior")_ , in which he introduced a [theoretical neural structure](https://en.wikipedia.org/wiki/Hebbian_theory "Hebbian
ml.md_0_189
"Hebbian theory") formed by certain interactions among [nerve cells](https://en.wikipedia.org/wiki/Nerve_cells "Nerve cells").[[13]](https://en.wikipedia.org/wiki/Machine_learning#cite_note-13) Hebb's model of [neurons](https://en.wikipedia.org/wiki/Neuron "Neuron") interacting with one another set a groundwork for how AIs and machine learning
ml.md_0_190
for how AIs and machine learning algorithms work under nodes, or [artificial neurons](https://en.wikipedia.org/wiki/Artificial_neuron "Artificial neuron") used by computers to communicate data.[[12]](https://en.wikipedia.org/wiki/Machine_learning#cite_note-WhatIs-12) Other researchers who have studied human [cognitive
ml.md_0_191
who have studied human [cognitive systems](https://en.wikipedia.org/wiki/Cognitive_systems_engineering "Cognitive systems engineering") contributed to the modern machine learning technologies as well, including logician [Walter Pitts](https://en.wikipedia.org/wiki/Walter_Pitts "Walter Pitts") and [Warren
ml.md_0_192
"Walter Pitts") and [Warren McCulloch](https://en.wikipedia.org/wiki/Warren_Sturgis_McCulloch "Warren Sturgis McCulloch"), who proposed the early mathematical models of neural networks to come up with [algorithms](https://en.wikipedia.org/wiki/Algorithm "Algorithm") that mirror human thought
ml.md_0_193
that mirror human thought processes.[[12]](https://en.wikipedia.org/wiki/Machine_learning#cite_note-WhatIs-12)
ml.md_0_194
By the early 1960s, an experimental "learning machine" with [punched tape](https://en.wikipedia.org/wiki/Punched_tape "Punched tape") memory, called Cybertron, had been developed by [Raytheon Company](https://en.wikipedia.org/wiki/Raytheon_Company "Raytheon Company") to analyse [sonar](https://en.wikipedia.org/wiki/Sonar "Sonar") signals,
ml.md_0_195
"Sonar") signals, [electrocardiograms](https://en.wikipedia.org/wiki/Electrocardiography "Electrocardiography"), and speech patterns using rudimentary [reinforcement learning](https://en.wikipedia.org/wiki/Reinforcement_learning "Reinforcement learning"). It was repetitively "trained" by a human operator/teacher to recognise patterns and equipped
ml.md_0_196
to recognise patterns and equipped with a "[goof](https://en.wikipedia.org/wiki/Goof "Goof")" button to cause it to reevaluate incorrect decisions.[[14]](https://en.wikipedia.org/wiki/Machine_learning#cite_note-14) A representative book on research into machine learning during the 1960s was Nilsson's book on Learning Machines, dealing mostly with
ml.md_0_197
Machines, dealing mostly with machine learning for pattern classification.[[15]](https://en.wikipedia.org/wiki/Machine_learning#cite_note-15) Interest related to pattern recognition continued into the 1970s, as described by Duda and Hart in 1973.[[16]](https://en.wikipedia.org/wiki/Machine_learning#cite_note-16) In 1981 a report was given on using
ml.md_0_198
1981 a report was given on using teaching strategies so that an [artificial neural network](https://en.wikipedia.org/wiki/Artificial_neural_network "Artificial neural network") learns to recognise 40 characters (26 letters, 10 digits, and 4 special symbols) from a computer
ml.md_0_199
4 special symbols) from a computer terminal.[[17]](https://en.wikipedia.org/wiki/Machine_learning#cite_note-17)
ml.md_0_200
[Tom M. Mitchell](https://en.wikipedia.org/wiki/Tom_M._Mitchell "Tom M. Mitchell") provided a widely quoted, more formal definition of the algorithms studied in the machine learning field: "A computer program is said to learn from experience _E_ with respect to some class of tasks _T_ and performance measure _P_ if its performance at tasks in _T_
ml.md_0_201
if its performance at tasks in _T_ , as measured by _P_ , improves with experience _E_."[[18]](https://en.wikipedia.org/wiki/Machine_learning#cite_note-Mitchell-1997-18) This definition of the tasks in which machine learning is concerned offers a fundamentally [operational definition](https://en.wikipedia.org/wiki/Operational_definition
ml.md_0_202
"Operational definition") rather than defining the field in cognitive terms. This follows [Alan Turing](https://en.wikipedia.org/wiki/Alan_Turing "Alan Turing")'s proposal in his paper "[Computing Machinery and Intelligence](https://en.wikipedia.org/wiki/Computing_Machinery_and_Intelligence "Computing Machinery and Intelligence")", in which the
ml.md_0_203
and Intelligence")", in which the question "Can machines think?" is replaced with the question "Can machines do what we (as thinking entities) can do?".[[19]](https://en.wikipedia.org/wiki/Machine_learning#cite_note-19)
ml.md_0_204
Modern-day machine learning has two objectives. One is to classify data based on models which have been developed; the other purpose is to make predictions for future outcomes based on these models. A hypothetical algorithm specific to classifying data may use computer vision of moles coupled with supervised learning in order to train it to
ml.md_0_205
learning in order to train it to classify the cancerous moles. A machine learning algorithm for stock trading may inform the trader of future potential predictions.[[20]](https://en.wikipedia.org/wiki/Machine_learning#cite_note-20)
ml.md_0_206
## Relationships to other fields [[edit](https://en.wikipedia.org/w/index.php?title=Machine_learning&action=edit&section=2 "Edit section: Relationships to other fields")] ### Artificial intelligence [[edit](https://en.wikipedia.org/w/index.php?title=Machine_learning&action=edit&section=3 "Edit section: Artificial intelligence")]
ml.md_0_207
[![](https://upload.wikimedia.org/wikipedia/commons/thumb/1/1b/AI_hierarchy.svg/250px-AI_hierarchy.svg.png)](https://en.wikipedia.org/wiki/File:AI_hierarchy.svg)Machine learning as subfield of AI[[21]](https://en.wikipedia.org/wiki/Machine_learning#cite_note-journalimcms.org-21)
ml.md_0_208
As a scientific endeavour, machine learning grew out of the quest for [artificial intelligence](https://en.wikipedia.org/wiki/Artificial_intelligence "Artificial intelligence") (AI). In the early days of AI as an [academic discipline](https://en.wikipedia.org/wiki/Discipline_\(academia\) "Discipline \(academia\)"), some researchers were interested
ml.md_0_209
some researchers were interested in having machines learn from data. They attempted to approach the problem with various symbolic methods, as well as what were then termed "[neural networks](https://en.wikipedia.org/wiki/Artificial_neural_network "Artificial neural network")"; these were mostly
ml.md_0_210
network")"; these were mostly [perceptrons](https://en.wikipedia.org/wiki/Perceptron "Perceptron") and [other models](https://en.wikipedia.org/wiki/ADALINE "ADALINE") that were later found to be reinventions of the [generalised linear models](https://en.wikipedia.org/wiki/Generalised_linear_model "Generalised linear model") of
ml.md_0_211
"Generalised linear model") of statistics.[[22]](https://en.wikipedia.org/wiki/Machine_learning#cite_note-22) [Probabilistic reasoning](https://en.wikipedia.org/wiki/Probabilistic_reasoning "Probabilistic reasoning") was also employed, especially in [automated medical diagnosis](https://en.wikipedia.org/wiki/Automated_medical_diagnosis "Automated
ml.md_0_212
"Automated medical diagnosis").[[23]](https://en.wikipedia.org/wiki/Machine_learning#cite_note-aima-23): 488
ml.md_0_213
However, an increasing emphasis on the [logical, knowledge-based approach](https://en.wikipedia.org/wiki/Symbolic_AI "Symbolic AI") caused a rift between AI and machine learning. Probabilistic systems were plagued by theoretical and practical problems of data acquisition and
ml.md_0_214
problems of data acquisition and representation.[[23]](https://en.wikipedia.org/wiki/Machine_learning#cite_note-aima-23): 488 By 1980, [expert systems](https://en.wikipedia.org/wiki/Expert_system "Expert system") had come to dominate AI, and statistics was out of favour.[[24]](https://en.wikipedia.org/wiki/Machine_learning#cite_note-changing-24)
ml.md_0_215
Work on symbolic/knowledge-based learning did continue within AI, leading to [inductive logic programming](https://en.wikipedia.org/wiki/Inductive_logic_programming "Inductive logic programming")(ILP), but the more statistical line of research was now outside the field of AI proper, in [pattern
ml.md_0_216
field of AI proper, in [pattern recognition](https://en.wikipedia.org/wiki/Pattern_recognition "Pattern recognition") and [information retrieval](https://en.wikipedia.org/wiki/Information_retrieval "Information retrieval").[[23]](https://en.wikipedia.org/wiki/Machine_learning#cite_note-aima-23): 708–710, 755 Neural networks research had been
ml.md_0_217
Neural networks research had been abandoned by AI and [computer science](https://en.wikipedia.org/wiki/Computer_science "Computer science") around the same time. This line, too, was continued outside the AI/CS field, as "[connectionism](https://en.wikipedia.org/wiki/Connectionism "Connectionism")", by researchers from other disciplines including
ml.md_0_218
from other disciplines including [John Hopfield](https://en.wikipedia.org/wiki/John_Hopfield "John Hopfield"), [David Rumelhart](https://en.wikipedia.org/wiki/David_Rumelhart "David Rumelhart"), and [Geoffrey Hinton](https://en.wikipedia.org/wiki/Geoffrey_Hinton "Geoffrey Hinton"). Their main success came in the mid-1980s with the reinvention of
ml.md_0_219
mid-1980s with the reinvention of [backpropagation](https://en.wikipedia.org/wiki/Backpropagation "Backpropagation").[[23]](https://en.wikipedia.org/wiki/Machine_learning#cite_note-aima-23): 25
ml.md_0_220
Machine learning (ML), reorganised and recognised as its own field, started to flourish in the 1990s. The field changed its goal from achieving artificial intelligence to tackling solvable problems of a practical nature. It shifted focus away from the [symbolic approaches](https://en.wikipedia.org/wiki/Symbolic_artificial_intelligence "Symbolic
ml.md_0_221
"Symbolic artificial intelligence") it had inherited from AI, and toward methods and models borrowed from statistics, [fuzzy logic](https://en.wikipedia.org/wiki/Fuzzy_logic "Fuzzy logic"), and [probability theory](https://en.wikipedia.org/wiki/Probability_theory "Probability
ml.md_0_222
"Probability theory").[[24]](https://en.wikipedia.org/wiki/Machine_learning#cite_note-changing-24)
ml.md_0_223
### Data compression [[edit](https://en.wikipedia.org/w/index.php?title=Machine_learning&action=edit&section=4 "Edit section: Data compression")]
ml.md_0_224
This section is an excerpt from [Data compression § Machine learning](https://en.wikipedia.org/wiki/Data_compression#Machine_learning "Data compression").[[edit](https://en.wikipedia.org/w/index.php?title=Data_compression&action=edit#Machine_learning)]
ml.md_0_225
There is a close connection between machine learning and compression. A system that predicts the [posterior probabilities](https://en.wikipedia.org/wiki/Posterior_probabilities "Posterior probabilities") of a sequence given its entire history can be used for optimal data compression (by using [arithmetic
ml.md_0_226
compression (by using [arithmetic coding](https://en.wikipedia.org/wiki/Arithmetic_coding "Arithmetic coding") on the output distribution). Conversely, an optimal compressor can be used for prediction (by finding the symbol that compresses best, given the previous history). This equivalence has been used as a justification for using data
ml.md_0_227
as a justification for using data compression as a benchmark for "general
ml.md_0_228
as a benchmark for "general intelligence".[[25]](https://en.wikipedia.org/wiki/Machine_learning#cite_note-Data_compression_Mahoney-25)[[26]](https://en.wikipedia.org/wiki/Machine_learning#cite_note-Data_compression_Market_Efficiency-26)[[27]](https://en.wikipedia.org/wiki/Machine_learning#cite_note-Data_compression_Ben-Gal-27)
ml.md_0_229
An alternative view can show compression algorithms implicitly map strings into implicit [feature space vectors](https://en.wikipedia.org/wiki/Feature_space_vector "Feature space vector"), and compression-based similarity measures compute similarity within these feature spaces. For each compressor C(.) we define an associated vector space ℵ,
ml.md_0_230
an associated vector space ℵ, such that C(.) maps an input string x, corresponding to the vector norm ||~x||. An exhaustive examination of the feature spaces underlying all compression algorithms is precluded by space; instead, feature vectors chooses to examine three representative lossless compression methods, LZW, LZ77, and
ml.md_0_231
methods, LZW, LZ77, and PPM.[[28]](https://en.wikipedia.org/wiki/Machine_learning#cite_note-Data_compression_ScullyBrodley-28)
ml.md_0_232
According to [AIXI](https://en.wikipedia.org/wiki/AIXI "AIXI") theory, a connection more directly explained in [Hutter Prize](https://en.wikipedia.org/wiki/Hutter_Prize "Hutter Prize"), the best possible compression of x is the smallest possible software that generates x. For example, in that model, a zip file's compressed size includes both the
ml.md_0_233
compressed size includes both the zip file and the unzipping software, since you can not unzip it without both, but there may be an even smaller combined form.
ml.md_0_234
Examples of AI-powered audio/video compression software include [NVIDIA Maxine](https://en.wikipedia.org/wiki/NVIDIA_Maxine "NVIDIA Maxine"), AIVC.[[29]](https://en.wikipedia.org/wiki/Machine_learning#cite_note-29) Examples of software that can perform AI-powered image compression include [OpenCV](https://en.wikipedia.org/wiki/OpenCV "OpenCV"),
ml.md_0_235
"OpenCV"), [TensorFlow](https://en.wikipedia.org/wiki/TensorFlow "TensorFlow"), [MATLAB](https://en.wikipedia.org/wiki/MATLAB "MATLAB")'s Image Processing Toolbox (IPT) and High-Fidelity Generative Image Compression.[[30]](https://en.wikipedia.org/wiki/Machine_learning#cite_note-30)
ml.md_0_236
In [unsupervised machine learning](https://en.wikipedia.org/wiki/Unsupervised_machine_learning "Unsupervised machine learning"), [k-means clustering](https://en.wikipedia.org/wiki/K-means_clustering "K-means clustering") can be utilized to compress data by grouping similar data points into clusters. This technique simplifies handling extensive
ml.md_0_237
simplifies handling extensive datasets that lack predefined labels and finds widespread use in fields such as [image compression](https://en.wikipedia.org/wiki/Image_compression "Image compression").[[31]](https://en.wikipedia.org/wiki/Machine_learning#cite_note-31)
ml.md_0_238
Data compression aims to reduce the size of data files, enhancing storage efficiency and speeding up data transmission. K-means clustering, an unsupervised machine learning algorithm, is employed to partition a dataset into a specified number of clusters, k, each represented by the [centroid](https://en.wikipedia.org/wiki/Centroid "Centroid") of
ml.md_0_239
"Centroid") of its points. This process condenses extensive datasets into a more compact set of representative points. Particularly beneficial in [image](https://en.wikipedia.org/wiki/Image_processing "Image processing") and [signal processing](https://en.wikipedia.org/wiki/Signal_processing "Signal processing"), k-means clustering aids in data
ml.md_0_240
k-means clustering aids in data reduction by replacing groups of data points with their centroids, thereby preserving the core information of the original data while significantly decreasing the required storage space.[[32]](https://en.wikipedia.org/wiki/Machine_learning#cite_note-32)
ml.md_0_241
[Large language models](https://en.wikipedia.org/wiki/Large_language_model "Large language model") (LLMs) are also efficient lossless data compressors on some data sets, as demonstrated by [DeepMind](https://en.wikipedia.org/wiki/DeepMind "DeepMind")'s research with the Chinchilla 70B model. Developed by DeepMind, Chinchilla 70B effectively
ml.md_0_242
Chinchilla 70B effectively compressed data, outperforming conventional methods such as [Portable Network Graphics](https://en.wikipedia.org/wiki/Portable_Network_Graphics "Portable Network Graphics") (PNG) for images and [Free Lossless Audio Codec](https://en.wikipedia.org/wiki/Free_Lossless_Audio_Codec "Free Lossless Audio Codec") (FLAC) for
ml.md_0_243
Lossless Audio Codec") (FLAC) for audio. It achieved compression of image and audio data to 43.4% and 16.4% of their original sizes, respectively. There is, however, some reason to be concerned that the data set used for testing overlaps the LLM training data set, making it possible that the Chinchilla 70B model is only an efficient compression
ml.md_0_244
is only an efficient compression tool on data it has already been trained on.[[33]](https://en.wikipedia.org/wiki/Machine_learning#cite_note-33)[[34]](https://en.wikipedia.org/wiki/Machine_learning#cite_note-34)
ml.md_0_245
### Data mining [[edit](https://en.wikipedia.org/w/index.php?title=Machine_learning&action=edit&section=5 "Edit section: Data mining")]
ml.md_0_246
Machine learning and [data mining](https://en.wikipedia.org/wiki/Data_mining "Data mining") often employ the same methods and overlap significantly, but while machine learning focuses on prediction, based on _known_ properties learned from the training data, data mining focuses on the
ml.md_0_247
data, data mining focuses on the [discovery](https://en.wikipedia.org/wiki/Discovery_\(observation\) "Discovery \(observation\)") of (previously) _unknown_ properties in the data (this is the analysis step of [knowledge discovery](https://en.wikipedia.org/wiki/Knowledge_discovery "Knowledge discovery") in databases). Data mining uses many machine
ml.md_0_248
Data mining uses many machine learning methods, but with different goals; on the other hand, machine learning also employs data mining methods as "[unsupervised learning](https://en.wikipedia.org/wiki/Unsupervised_learning "Unsupervised learning")" or as a preprocessing step to improve learner accuracy. Much of the confusion between these two
ml.md_0_249
of the confusion between these two research communities (which do often have separate conferences and separate journals, [ECML PKDD](https://en.wikipedia.org/wiki/ECML_PKDD "ECML PKDD") being a major exception) comes from the basic assumptions they work with: in machine learning, performance is usually evaluated with respect to the ability to
ml.md_0_250
with respect to the ability to _reproduce known_ knowledge, while in knowledge discovery and data mining (KDD) the key task is the discovery of previously _unknown_ knowledge. Evaluated with respect to known knowledge, an uninformed (unsupervised) method will easily be outperformed by other supervised methods, while in a typical KDD task,
ml.md_0_251
while in a typical KDD task, supervised methods cannot be used due to the unavailability of training data.
ml.md_0_252
Machine learning also has intimate ties to [optimisation](https://en.wikipedia.org/wiki/Optimisation "Optimisation"): Many learning problems are formulated as minimisation of some [loss function](https://en.wikipedia.org/wiki/Loss_function "Loss function") on a training set of examples. Loss functions express the discrepancy between the
ml.md_0_253
the discrepancy between the predictions of the model being trained and the actual problem instances (for example, in classification, one wants to assign a [label](https://en.wikipedia.org/wiki/Labeled_data "Labeled data") to instances, and models are trained to correctly predict the preassigned labels of a set of
ml.md_0_254
the preassigned labels of a set of examples).[[35]](https://en.wikipedia.org/wiki/Machine_learning#cite_note-35)
ml.md_0_255
### Generalization [[edit](https://en.wikipedia.org/w/index.php?title=Machine_learning&action=edit&section=6 "Edit section: Generalization")]
ml.md_0_256
Characterizing the generalisation of various learning algorithms is an active topic of current research, especially for [deep learning](https://en.wikipedia.org/wiki/Deep_learning "Deep learning") algorithms. ### Statistics [[edit](https://en.wikipedia.org/w/index.php?title=Machine_learning&action=edit&section=7 "Edit section: Statistics")]
ml.md_0_257
Machine learning and [statistics](https://en.wikipedia.org/wiki/Statistics "Statistics") are closely related fields in terms of methods, but distinct in their principal goal: statistics draws population [inferences](https://en.wikipedia.org/wiki/Statistical_inference "Statistical inference") from a
ml.md_0_258
"Statistical inference") from a [sample](https://en.wikipedia.org/wiki/Sample_\(statistics\) "Sample \(statistics\)"), while machine learning finds generalisable predictive patterns.[[36]](https://en.wikipedia.org/wiki/Machine_learning#cite_note-36) According to [Michael I. Jordan](https://en.wikipedia.org/wiki/Michael_I._Jordan "Michael I.
ml.md_0_259
"Michael I. Jordan"), the ideas of machine learning, from methodological principles to theoretical tools, have had a long pre-history in statistics.[[37]](https://en.wikipedia.org/wiki/Machine_learning#cite_note-mi_jordan_ama-37) He also suggested the term [data science](https://en.wikipedia.org/wiki/Data_science "Data science") as a placeholder
ml.md_0_260
"Data science") as a placeholder to call the overall field.[[37]](https://en.wikipedia.org/wiki/Machine_learning#cite_note-mi_jordan_ama-37)
ml.md_0_261
Conventional statistical analyses require the a priori selection of a model most suitable for the study data set. In addition, only significant or theoretically relevant variables based on previous experience are included for analysis. In contrast, machine learning is not built on a pre-structured model; rather, the data shape the model by
ml.md_0_262
the data shape the model by detecting underlying patterns. The more variables (input) used to train the model, the more accurate the ultimate model will be.[[38]](https://en.wikipedia.org/wiki/Machine_learning#cite_note-38)
ml.md_0_263