The map is centered on the topic of inference, which consists of using rules of inference to form conclusions from a premise set. To fully understand the concept of inference, one must understand both of the components that enable an inference to be made: namely rules of inference and premise sets.

taupesalmonInternet και Εφαρμογές Web

21 Οκτ 2013 (πριν από 3 χρόνια και 9 μήνες)

86 εμφανίσεις

The map is centered on the topic of inference, which consists of using rules of inference to form
conclusions from a premise set. To fully understand the concept of inference, one must understand
both of the components that enable an inference to be
made: namely rules of inference and premise
sets.


Rules of inference can be described as logical transformations used to produce conclusions from
the premise set. These rules almost always consist of fundamental rules from the study of logic, such as
Mod
us Ponens. This states that if the truth of premise A implies that premise B is true, and it can be
shown (or it is given) that premise A is true, then premise B is always true. Rules such as this form the
framework of formal logic that most forms of inf
erence are built upon.


In addition to formal logical rules, a rule set for a specific instance of inference can include rules
pertaining to specific qualities of the data you are working with that build on the formal inference rules.
These most often t
ake the form of providing context for a formal rule of logic. For instance, the
inference rule “if John is an employee, he gets paid” would expand upon Modus Ponens to provide the
condition that must be satisfied for us to conclude that John gets paid.


A
s with most cases, inference rules must be in some sort of standardized format to be utilized
by computerized inference engines. These form
ats, called ontology languages, provide the structure
necessary for efficient software
processing of inference rules
. Many widely used ontology languages
exist today, such as OWL, which uses either Resource Description Framework (RDF) or XML to
structuralize inference rules, and DAML + OIL, which is an ontology language developed by DARPA to
structuralize inference rul
es.


Another product of standardizing inference rules with ontology languages is what is known as
the semantic web. The semantic web is defined as the collection of structured data, including ontology
languages, that is present on the World Wide Web. Adv
anced search engines, such as Swoogle, were
created to enable researchers to find specific sets of inference rules that have been publicly shared.



The other component necessary for inference is the premise set. This consists of all the “given”
data that

is assumed true before starting the inference process. It is this data that the inference rules
are applied against.
One of the most popular ways to generate a premise set is through data mining,
which describes a methodology for collecting and structur
ing data.

In its purest sense, data mining is
the collection of raw data into some more easily manageable form. There
also exist certain
specializations of data mining, one of which is known as knowledge gathering. This consists of searching
large volum
es of data for patterns, which are known as knowledge. While in some cases, the discovery
of the patterns themselves is enough to satisfy the purpose of the data mining, the patterns can also be
used as a premise set in order to draw further conclusions a
bout the data with inference.


Another form of data mining, known as data warehousing, consists of the archiving of data into
a structured repository of information. This technique is used for inference more often than knowledge
gathering due to the fact
that its sole purpose is to condense data into a usable form, without
identifying any patterns or drawing conclusions. Data warehousing is also commonly used to prepare
data for analysis by reporting software, which is differentiated from inference by the

fact that most
reporting software does not draw new conclusions from the processed data, and only statistical
information, such as averages and means.


With a collection of inference rules and a premise set, it is possible to form conclusions from
data.
This is usually accomplished by using inference rules to form intermediate conclusions about the
data, which are then added to the premise set. This process is repeated until either the desired
information is achieved, or no further conclusions can be mad
e.


It is with these basic concepts that inference is put to a number of practical uses. The first of
these is predictive analytics, which analyzes current and historic data to make conclusions about future
events. The most common uses of this field ar
e stock market forecasting, election prediction, and
market analytics, where analysts predict the success of individual businesses or markets. While the
majority of this field can be labeled as prediction, they do make use of inference in order to justify

these
predictions. An example of this would be using a premise set of market conditions and political
conditions for a certain year to attempt to determine the impact of certain political polices on the
market. If inference can be used to conclude that
a certain political action has a negative impact on the
market, then analysts will predict a future downturn of the market if another politician makes the same
action.


Another applied, albeit malicious, use of inference is the inference attack. This uses

the
methodology of inference to derive classified, protected, or privileged data from a premise set that is
considered safe to divulge.
A form of inference attacks, known as statistical inference attacks, can be as
simple as reversing statistical calcula
tions given by reporting software, such as salary averages, in order
to conclude the individual salaries of an employee. In this case, the premise set is not only considered
safe to divulge, but is actually created for the sole reason of being divulged to

users.


Inference attacks are of specific interest to security researchers due to the fact that they require
no errors in code or holes in perimeter defense to execute. In most cases, an inference attack manifests
itself as seemingly normal user requests
.

This makes mitigation of these attacks a much more involved
process than traditional security. One must consider all the ways in which secrets can be derived from
the data presented to their users. This includes not only information divulged by the ho
lders of the
data, but also public data ava
ilable by other means, such as domain registration records obtained with
Whois

or shareholder earnings reports.


A third area where inference is applied is the realm of artificial intelligence, specifically the
cr
eation of expert systems. An expert system is software that attempts to reproduce the performance
of an expert human in a specific field. A common example of this are online “font finder” sites, where
an expert system can determine the name of a specific

font by asking the user a series of questions
pertaining to the attributes of certain letters.


An expert system contains a rule base and an inference engine. The rule base functions as the
rules of inference, and consists of ontologies governing the beh
avior of the expert system.

The
inference engine, on the other hand, is the part of the software that actually uses the rule base to draw
conclusion from the premise set, which in this case is supplied by the user of the system. Since the
premise set of
an expert system is provided one item at a time, the inference engine must be able to
discern which piece of data would be the most useful in arriving at a conclusion. This will be the piece of
data the expert system next asks for.


There are two types of

inference engines used in expert systems: forward chaining engines and
backward chaining engines. In a forward chaining engine, the engine takes the given data and draws as
many conclusions as it can from said data. It then checks to see if any of the
conclusions are the same as
the goal of the expert system, which is the conclusion desired by the user. If none of the conclusions are
equal to the goal, they are added to the premise set and the process is repeated. Because of this,
forward chaining inf
erence engines are also known as data driven engines.


On the other hand, backward chaining engines first consider the goal of the expert system. It
then determines what data would be necessary to prove the goal with the current inference rule set. If
th
is data is present in the premise set. If the data is present, then the expert system arrives at a
conclusion and is finished. If not, the data that was not present in the premise set is then added to the
list of goals, and the process is repeated until
either a conclusion is arrived at, or all the data has been
processed.