AIMA Chapter 12: Knowledge Representation

General Purpose Ontology

1. should be applicable in any special purpose domain, with the addition of domain specific axioms. 2. give a demanding enough domain, different areas of knowledge should be unified.

Count noun

A noun that is countable, and in Standard American English, is preceeded by an "a" or "an". Ex: an aardvark, a potatoe .. etc

Qualitative Physics

A sub-field of AI that investigates how to reason about physical systems without plunging into detailed equations and numerical simulations.

Procedural Attachment

A technique whereby a query about (or sometimes an assertion of) a certain relation results in a call to a special procedure designed for that relation rather than a general inference algorithm.

Moment

A time interval that has 0 duration

Iheritence

Any object in a category that is within an outer category, belongs to both the inner and outer category

Categories

Breaking down objects into groups of categories is vital to ontology with many applications, such as category membership.

Process Categories / Liquid Event Categories

Categories where a subinterval of an event still counts as an event in the category. ex: the last 20 minutes of a flight still belongs to the flight category. (e ∈ Processes) ∧ Happens(e, (t1, t4)) ∧ (t1 < t2 < t3 < t4) ⇒ Happens(e, (t2, t3))

Subsumption

Checking if one category is a subset of another by comparing their definitions. The principal inference t asks for description logics.

Classification

Checking whether an object belongs to a category.

Justification-based truth maintenance system (JTMS)

Each sentence in the KB is annotated with a justification consisting of the set of sentences from which it was inferred. When a sentence P is retracted, it will delete every sentence where P is in every justification for the sentence. Rather than removing any sentence from the KB, JTMS only marks sentences as out, if it's justifications are retracted.

Upper Ontology

Framework of abstract concepts. More specific concepts come underneath the general ones.

Logical Omniscience

If an agent knows a set of axioms, then it knows all consequences of those axioms. Logical Omniscience is assumed by modal logic, and this can be problematic.

Possible Worlds

In Modal logic, there exists multiple possible worlds rather than just one world with truth.

Ambiguity

One name for two or more different categories.

Intrinsic Properties

Properties that belong to the very substance of the object, rather than the object as a whole ex: density, boiling point, etc.

Description Logics

Provide a formal language for constructing and combining category definitions and efficient algorithms for deciding subset and super-set relationships between categories. Notations that are designed to make it easier to describe definitions and properties of categories.

Ontological Engineering

Representation of abstract concepts, such as events, objects etc.

Default Values and Overriding

Semantic networks have a strong ability to assert default values. Categories are given default values, that stand unless overridden. Ex: all people are given the default value of two legs, but when we say that John has one leg, this does not become a contradiction, it is just overridden.

General Categories

Stuff: the most general substance category, and has no intrinsic properties Thing: the most general discrete category, and has no extrinsic properties

Truth Maintenance Systems (TMS)

Systems designed to maintain the truth of the KB when a certain sentence is revised.

Explenations

TMSs are able to generate explenations, or a set of sentences E that justify the sentence P. Although not all sentences in E have to be true, they can also just be assumed. Ideally E should be minimal, that is no subset of E should be able to alone explain P.

Units Function

Takes a number as an argument

Stuff

The generic name for the portion of reality that defies individuation. Ex: butter cannot be broken up into 5 butters, a tub of butter is still a butter object even if it's half empty. things on the other hand do not have this ambiguity. If you cut an arrdvark in half, you do not have two arrdvarks.

Referntial Opacity

The opposite of referential transparency, where the terms do in fact matter. This is usefull since not all agents know which terms are co-referential.

Belief Revision

The removal of facts that were inferred but turned out to be wrong. Can be rather tricky as other facts may be inferred by the fact that is being removed.

Time Functions

Time(t) : delivers the point on the time scale for a moment Begin(moment): returns the earliest moment in it's time interval End(moment): returns the latest moment in the event's time scale Duration(event): gives the difference between the end time and start time Date(hours, minutes, seconds, day, month, year): returns a time point

Reification

To make a category or event into an object itself as opposed to using only predicates to represent an object being a member of a category.

Assumption-based Truth Maintenance System (ATMS)

a TMS where each sentence has a set of assumptions that would make the sentence true. This TMS is particularly useful for representing hypothetical worlds.

Natural Kind

a category that does not have strict definitions and many exceptions exist. This poses a problem, so often times a separate category defines what is typical of the category, but not necessary.

Partition

a disjoint exhaustive decomposition

Event Calculus

a formalism based on points of time rather than on situations T(f, t) Fluent f is true at time t Happens(e, i) Event e happens over the time interval i Initiates(e, f, t) Even e causes fluent f to start to hold at time t Terminates(e, f, t) event e causes fluent f to cease to hold at time t Clipped(f, i) Fluent f ceased to be true at some point during time interval i Restored(f, i) Fluent f becomes true sometime during interval i

Default Logic

a formalism in which default ruels can be written to generate contingent nonmonotonic conclusions. Ex: bird(x) : files(x)/Files(x) in general: P: j1, j2,....., jn/C where P is the prerequisite, C is the conclusion and ji are the justifications. If any of the justifications are false, then the conclusion cannot be drawn.

Prioritized Circumscription

a formalism that gives preference to minimize certain circumscriptions over others. For example: we may want to minimize abonormal1 over minimization of abnormal2

Existential Graphs

a graphical notation of nodes and edges

Circumscription

a more precise version of the closed world assumption where certain predicates are assumed to be "as false as possible" that is false for every object except those for which they are known to be true. any predicate that is circumscribed is assumed to be false unless specifically noted as true

Mass Nouns

a noun that doesn't have clear individuation. Ex: butter, water, energy etc.

Wrapper

a program used to extract certain information from a page.

Temporal Substances vs Spatial Substances

a substance is an object who's parts still makeup the object itself. A temporal substance may be broken up by time, and still be in the same category. A spatial substance may be broken up my space or volume, and still be in the same category. ex: temporal substance: flight; Spatial substance: butter

composite Objects

an object can be described as a part of another object. a composite object is the sum of the masses of the parts.

Propositional Attitudes

attitudes an agent can have toward mental objects, such as: Believes, Knows, Wants, Intends, and Informs.

Consistency

checks whether the membership criteria are logically satisfiable.

Measures

cost, height, mass, etc. Quantitative characteristics that belong to objects. Measures do not have to be numbers, but should be able to be ordered.

Logical Minimization

defining an object as the smallest one satisfying certain conditions.

Individuation

division into distinct objects.

Discrete Events

events with a definite structure

Model Preference Logics

in these logics, a a sentence is entailed (with default status) if it is true in all preferred models of t he KB as opposed to the requirement of truth in ALL possible models like it is in classical logic.

Exhaustive Decomposition

indicates that all objects in a category must belong to one of the given subcategories. Ex: all humans need to belong to either male or female subcategory.

Referential Transparency

it doesn't matter what term a logic uses to refer to an object, what matter is the object that the term names.

Monmonotonicity / nonmonotonic logic

logics that have been devised with modified notions of truth and entailment to capture the exceptions where a set of beliefs does not necessarily grow with each new assertion. Two extensivley studied such logics are: Circumscription and default logic

Modal Logic

modal logic allows for representation of multiple modes, as opposed to regular logic that is concerned with one mode only: the truth. Modal syntax is the same as FOL but has it's own additional operators.

Semantic Networks

one of the systems designed specially for organizing and reasoning with categories. A semantic network provides graphical aids for visualizing a knowledge base and efficient algorithms for inferring properties of an object on the basis of its category membership.

Extrinisic Properties

properties that are specific to certain objects. Ex: weight, height, shape etc.

Accessibility Relations

relations that connect the possible worlds in modal logic. Acc(Ka, w0, w1) means that w1 is accessible from w0 with respect to the modal operator Ka if everything in w1 is consistent with what a knows in w0.

Bunch

similar to a set, but a bunch is a composite object made of the objects in the bunch. ex: BunchOf({Apple1, Apple2, Apple3})

Synonymy

two names for the same category. Such as "laptop Computers" and "laptops"

Disjoint

two or more categories are disjoint if no single object can be in both categories. ex: females and males

Multiple Inheritance

when a category can be a subset of more than one other category, or an object can belong to more than one category.