What is Machine Learning

 The term machine learning (ML) as a part of artificial intelligence is ubiq-

uitous nowadays. The term is used for a wide number of various appli-

cations and methods that deal with the “generation of knowledge from

experience”.

The well-known US computer scientist Tom Mitchell defines machine

learning as follows:

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, as measured by P, improves with experience E (Mitchell 1997).

An illustrative example of this would be a chess computer program that

improves its performance (P) in playing chess (the task T) by experience

(E), by playing as many games as possible (even against itself ) and analysing

them (Mitchell 1997).

Machine learning is not a fundamentally new approach for machines to

generate “knowledge” from experience. Machine learning technology was

used to filter out junk e-mails a long time ago. Whilst spam filters that tack-

led the problem with the help of knowledge modelling had to constantly be

adapted manually, ML algorithms learn more with each e-mail and are able

to autonomously adapt their performance accordingly.

Besides the fields of responsibility defined in the previous section, with

machine learning, different ways of learning are differentiated from each

other. 

The most common will be discussed in the following:

Supervised Learning

Supervised learning proceeds within clearly defined limits. Besides the actual

data set, the right possible answers are already known. The supervised learn-

ing methods are meant to reveal the relationship between input and out-

put data. These methods are used for tasks in the fields of classification as

well as regression analyses. Regression is about predicting the results within

a continuous output, which means that an attempt is made to allocate input

variables to a continuous function. With the classification, in contrast, an

attempt is made to predict results in a discreet output, i.e. allocate input var-

iables to discreet categories.

The forecast of property prices, for example, based on the size of the

houses, would be an example for a regression problem. If, instead of that, we

forecast whether a house will cost more or less than a certain price depend-

ing on its size, that would be a categorisation, where the house would be

placed in two discreet categories according to the price.

Unsupervised Learning

In contrast to supervised learning, with unsupervised learning, the system

is not given target values labelled in advance. It is meant to autonomously

identify commonalities in the data sets and then form clusters or compro-

mise the data. As a rule, it is about discovering patterns in data that humans

are unaware of.

Unsupervised learning algorithms can, for example, be used for customer

or market segmentation or for clustering genes in genetic research, to reduce

the number of characteristic values. With the help of this compression, com-

puting could be faster afterwards without loss of data.

Reinforcement Learning

An alternative to unsupervised learning is provided by the models of rein-

forcement learning where learning patterns from nature are reproduced in

concepts. Through the combination of dynamic programming and super-

vised learning, problems that previously seemed to be unsolvable can be

solved. Differently to unsupervised learning, the system does not have an

ideal approach at the beginning of the learning phase. This has to be found

step by step by trial and error. Good approaches are rewarded and steps 

tending to be bad are sanctioned with penalisation. The system is able to

incorporate a multitude of environmental influences into the decisions

made and to respond to them. Reinforcement learning belongs to the field

of exploration learning, where a system autonomously, thus irrespective

of the rewards and penalties pointing in the right direction, has to find its

own solutions that can be clearly differentiated from those thought up by

humans. Reinforcement learning attracted a notable amount of attention

after the victory of Google DeepMind’s AlphaGo over Lee Sedol. The system

used applied deep reinforcement learning among others to improve its strat-

egy in simulated games against itself. Through reinforcement learning, arti-

ficial intelligences thus acquire the ability to find new approaches on their

own and to at least seemingly act intuitively