62
MP2-1 Default Reasoning and Theory Change G. Antoniou A.Nayak A. Ghose

Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

  • Upload
    others

  • View
    2

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-1

Default Reasoningand

Theory Change

G. Antoniou A.Nayak A. Ghose

Page 2: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-2

Part I

Introduction to Default Logic

Page 3: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-3

Nonmonotonic Reasoning — Motivation

How do you get to work? By bus!

Usually, I go to work by bus.

I walk to the bus stop and read:

No buses today, we’re on strike.

Now I have to take back my previous conclusion: nonmonotonicity.

Page 4: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-4

Incomplete information

Why not use the classical rule

goToWork ! ¬ strike " takeBus?

• Can I list all potential obstacles? (icy streets, being in a hurry etc.).

• I may not have the time or the resources to establish theconditions of the left hand: Incomplete information.

Page 5: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-5

Plausible conjectures

Intelligent systems need to make plausible conjectures, based on, say:

• Default rulesUsually I go to work by bus.

• IntrospectionIf the Rolling Stones were giving a concert in mycity tonight, I would have heard of that.

Page 6: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-6

Reasons for being interested in NMR

• Reasoning with incomplete information.

• Maintaining "competing" information within the same knowledge base.

• Compact representation of information.

• Pieces of information that are stable under changes.

Page 7: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-7

Default Logics — Overview

• The notion of a default

• Reiter’s default logic

- Syntax- Extensions- Properties

• Default logic variants

- Constrained default logic- Priorities among defaults

Page 8: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-8

The notion of a default

Defaults are rules of inference that can be applied if some informationis given, and some assumptions can be made.

Prototypical reasoningchild(X) : hasParents(X) / hasParents(X)

No-risk reasoningaccused(X) : innocent(X) / innocent(X)

Page 9: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-9

Defaults — Examples

Defaults in law

criminal(X) ! foreigner(X) : expel(X) / expel(X)

Exception:politicalRefugee(X) " ¬ expel(X)

Page 10: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-10

Defaults — Examples

Defaults in biology

Typically molluscs are shell-bearers.Cephalopods are molluscs.Cephalopods are not shell-bearers.

Expressed formally:

mollusc(X) : shellBearer(X) / shellBearer(X)cephalopod(X) " mollusc(X) ! ¬ shellBearer(X)

Page 11: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-11

Defaults — Examples

Closed World Assumption

true : ¬A / ¬A

for all ground facts A.

Page 12: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-12

Default logic — Types of knowledge

A default theory T consists of two kinds of knowledge:

• A set W of first order formulas called facts; certain information.

• A set D defaults; the assumptions that can be made.

Page 13: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-13

Defaults — Definition

A default d has the form

A : B1, . . . , Bn / C

with closed first order formulas A, Bi, C.

A is called the prerequisite pre(d), B1, . . ., Bn the justificationsjust(d), and C the consequent cons(d) of d.

Page 14: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-14

Defaults with variables

“Defaults” with free variables are read as schemes.

bird(X) : flies(X) / flies(X)

represents the set of defaults

bird(tweety) : flies(tweety) / flies(tweety)bird(sam) : flies(sam) / flies(sam). . .

Page 15: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-15

Extensions — Informal idea

Extensions are "world views" that are based on the given information(facts and defaults).

Extensions are obtained from the application of some defaults in D.They include always the certain information W.

Page 16: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-16

Interpretation of defaults

A : B1, . . ., Bn / C

If A is currently known,and if it is consistent to assume B1, . . ., Bn,then conclude C.

A : B1, . . ., Bn / C is applicable to a deductively closedset of formulas E iff A#E and ¬ B1$E, . . ., ¬ Bn$E.

Page 17: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-17

Extensions — Desirable properties

• They should be deductively closed under classical reasoning.

• They should be maximal: no more defaults can be applied.

Extensions represent maximal world views. In practice we may beinterested in portions of extensions (query evaluation).

Page 18: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-18

In-set and Out-set

P = (d0, d1, . . .) sequence of defaults from D without multiple occurrences.P[k] denotes the beginning part of P of length k.

• In(P) = Th(W % {cons(d) | d occurs in P}).

• Out(P) = {¬ B | B # just(d) for some d in P}.

Page 19: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-19

In-set and Out-set — Examples

aa : ¬ b / ¬ b [d1]b : c / c [d2]

Let P1 = (d1); In(P1) = Th({a, ¬ b}) and Out(P1) = {b}.

Let P2 = (d2, d1); In(P2) = Th({a, c, ¬ b}), Out(P2) = {¬ c, b}.

Page 20: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-20

Processes

Enforce that defaults can indeed be applied in the given order.P is called a process iff the defaults can be applied in the given order.

P is a process iff for every k such that P[k] is defined,dk is applicable to In(P[k]).

Page 21: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-21

Extensions — Definition

Given a process P we define the following:

• P is successful iff In(P) & Out(P) = ', otherwise it is failed.

• P is closed iff every default in D that is applicable to In(P) already occurs in P.

• E is an extension of a default theory T = (W,D) iff there is aclosed and successful process P of T such that E = In(P).

Page 22: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-22

Processes & Extensions — Example

aa : ¬ b / d [d1]true : c / b [d2]

• P1 = (d1) is successful but not closed.

• P2 = (d1, d2) is closed but failed.

• P3 = (d2) is a closed and successful process. Thus E = Th({a,b}) isan extension, in fact the only extension of the default theory.

Page 23: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-23

Process tree

• Nodes: (In, Out)

• Edges: (In, Out) " (In’, Out’) iff there is a defaultd = A : B1, . . ., Bn / C such that- d is applicable to In- In’ = Th(In % {C})- Out’ = Out % {¬B1, . . ., ¬Bn}

Page 24: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-24

Process tree

• Root: (Th(W), ')

• Paths starting at the root correspond to processes.

Page 25: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-25

A procedure for determining extensions

Given a default theory T.

1. Build the process tree of T.

2. Traverse the tree, and collect all closed & successful nodes(corresponding to closed & successful processes of T).

Page 26: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-26

Extensions — Example

true : a / ¬ a

Page 27: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-27

Extensions — Example

true : p / ¬ qtrue : q / r

Page 28: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-28

Extensions — Example

green : ¬ likesCars / ¬ likesCarsaaaMember : likesCars / likesCars

Page 29: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-29

Reiter’s definition of extensions

Let E and F be deductively closed sets of formulas, and D a set of defaults.

• A default A : B1, . . ., Bn / C is applicable to F with respect to belief set E iff A # F and ¬ B1 $ E, . . ., ¬ Bn $ E.

• F is closed under D w.r.t. belief set E iff, for every default d#Dthat is applicable to F w.r.t. E, cons(d)#F.

Page 30: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-30

Reiter’s definition of extensions

For T = (W,D) let (T(E) be the smallest set of formulas that• includes W,• is deductively closed, and• is closed under D w.r.t. E.

THEOREM [Antoniou 1996] E is an extension of T iff E = (T(E).

Note: Tradeoff between guessing and search.

Page 31: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-31

Default logic — Properties

• Existence of extensions

• Joint consistency of justifications

• Cumulativity

Page 32: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-32

Existence of extensions

We saw that it is not guaranteed. Is this undesirable?

• One may say no: user is responsible, as in programming

• The opposite view expects a logic to be more "fault tolerant".Different information sources (distributed systems,information superhighway).

Page 33: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-33

Ensuring existence of extensions

• Maintain the notion of extensions, but restrict the classes of defaulttheories considered.

• Modify the concept of an extension.

Page 34: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-34

Normal default theories

If we go the first way we may restrict attention to normal defaults:

A : B / B

[Reiter 1980] shows that normal default theories have alwaysextensions, essentially because all processes are successful.

Page 35: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-35

Normal defaults — Limitations

They may be unable to express interactions among defaults.

dropout(bill)dropout(X) : adult(X) / adult(X)adult(X) : employed(X) / employed(X)

Two extensions, but we would prefer one (which?)

In general, normal default theories are strictly less expressive thangeneral default theories.

Page 36: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-36

Semi-normal defaults

One possible solution:adult(X) : employed(X) ! ¬ dropout(X) / employed(X)

Semi-normal defaults:A : B ! C / C

[Etherington 1987] gives a sufficient condition for theexistence of extensions.

Page 37: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-37

Preference among extensions

Another way to ensure extensions is to use normal defaults, but add apriority relation which can model interactions among defaults.See Priorities among defaults later on.

Page 38: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-38

Modifications of the extension concept

To ensure the existence of extensions.

• Justified Default Logic [Lukaszewicz 1988]

• Constrained Default Logic [Schaub 1992]

Page 39: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-39

Joint consistency of justifications

Given the defaults

true : p / qtrue : ¬ p / r

there is a single extension, Th({q, r}).

Jumping to conclusions versus consistent set of beliefs.

Page 40: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-40

Approaches based on joint consistency

• Constrained Default Logic [Schaub 1992]

• Rational Default Logic [Mikitiuk & Truszczynski 1993]

Page 41: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-41

Joint consistency — Discussion

Is joint consistency a desirable property? Which is the right DL approach?

It depends very much on the problem at hand!

Reiter’s approach may lead to counterintuitive results:

true : usable(l) ! ¬ broken(l) / usable(l)true : usable(r) ! ¬ broken(r) / usable(r)broken(l) ) broken(r)

Page 42: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-42

Joint consistency — Discussion

But joint consistency may also lead to counterintuitive results.

travel : goodWeather / takeSwimSuittravel : badWeather / takeRainCoattravel¬ goodWeather ) ¬ badWeather

Here a cautious traveler would wish to apply both defaults, eventhough the assumptions contradict one another.

Page 43: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-43

Inference relations

In classical logic: M |– A. In default logic:

• Skeptical reasoningW |~D A iff A is included in all extensions of (W,D).

• Credulous reasoningW |~D A iff A is included in at least one extension of (W,D).

Page 44: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-44

Cumulativity

Ensures the safe use of lemmas (in the skeptical approach):

If W |~D A, then for all formulas B,

W |~D B * W%{A} |~D B.

Page 45: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-45

Default logic violates cumulativity

Consider the default theory T = (W,D) [Makinson 1989]:

true : p / p

p ) q : ¬ p / ¬ p

The only extension is Th({p}), so ' |~D p ) q.But if we add p ) q to T, then we get two extensions

Th({p})Th({¬ p, q})

p is not included in both extensions.

Page 46: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-46

Default Logics — Overview

• The notion of a default

• Reiter’s default logic

- Syntax- Extensions- Properties

• Default logic variants- Constrained default logic- Priorities among defaults

Page 47: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-47

Constrained default logic

JDL does not guarantee joint consistency of justifications.

true : p / qtrue : ¬ p / r

has the single modified extension Th({q, r}).

Constrained Default Logic [Schaub 1992] guarantees joint consistency.It does so by maintaining a consistent set of supporting beliefs.

Page 48: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-48

Default applicability in CDL

CDL adopts the idea of JDL not to run blindly into failure: "look ahead".

A default A : B1, . . ., Bn / C is applicable to deductively closed sets offormulas E (current knowledge) and Con (set of supporting beliefs) iff

A#E, and {B1, . . ., Bn, C} % Con is consistent.

Page 49: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-49

The supporting belief set

Let T = (W,D) be a default theory. Let P be a sequence of defaults(d0, d1, . . .) of defaults without multiple occurrences.

• Con(P) = Th(W % cons(P) % just(P)).

Exampletrue : p / q [d1]true : ¬ p / r [d2]

For the sequence P = (d1) we have Con(P) = Th({p, q}).

Page 50: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-50

Processes and extensions in CDL

• P is a constrained process iff dk is applicable to In(P[k]) and Con([k]), for all k such that P[k] is defined.

• P is a closed constrained process iff every default in D that isapplicable to In(P) and Con(P) already occurs in P.

• A pair (E, C) of deductively closed sets of formulas is aconstrained extension iff there is a closed constrained process Psuch that E = In(P) and C = Con(P).

Page 51: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-51

Constrained extensions — Example

pp : ¬ r / qp : r / r

Page 52: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-52

Constrained extensions — Example

true : usable(l) ! ¬ broken(l) / usable(l)true : usable(r) ! ¬ broken(r) / usable(r)

Page 53: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-53

Properties of CDL

THEOREM Every default theory has at least one constrained extension.

THEOREM Let E = In(P) for a closed and successful process P of T.If E % ¬ Out(P) is consistent then (E, Th(E % ¬ Out(P))) is a constrainedextension of T.

The converse is not true:true : p / ¬ p.

Page 54: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-54

Properties of CDL

THEOREM If (E,C) is a constrained extension of T then there is a modifiedextension F of T such that E + F.

Page 55: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-55

Comparison of default logics

THEOREM Let T be a normal default theory. The following statementsare equivalent:

(1) E is an extension of T.

(2) E is a modified extension of T.

(3) There is a set C such that (E,C) is a constrained extension of T.

Page 56: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-56

Priorities among defaults

• Default logics give overview of all possible extensions.

• But what if we wish to choose the most "important"(probable) possibilities?

- Medical diagnosis.

- Law.

• Technically achieved by introducing priorities among defaults.

Page 57: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-57

Default theories with priorities

penguin : ¬ flies / ¬ flies

bird : flies / flies

Give the first default higher priority than the second.

Total ordering among defaults: apply the default withthe highest priority that is applicable.

In the example above we get only one extension, Th({¬ flies}).

Page 58: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-58

Partial priority orders

Sometimes it is not reasonable to expect that the preference order be total.

conservative : taxCut ! ¬ spengingsCut / taxCut ! ¬ spengingsCut

conservative ! radical : taxCut ! spengingsCut / taxCut ! spengingsCut

socialDemocrat : ¬ taxCut ! ¬ spengingsCut / ¬ taxCut ! ¬ spengingsCut

Page 59: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-59

Partial priority orders

politicianpolitician : ¬ respected / ¬ respected [d1]politician : wellPaid / wellPaid [d2]¬ respected : ¬ wellPaid / ¬ wellPaid [d3]d2 < d3

Page 60: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-60

Consider “linearizations” of partial orders

According to all three total orderings which include <,

d1 << d2 << d3d2 << d1 << d3d2 << d3 << d1

we conclude wellPaid.

Page 61: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-61

References

G. Antoniou (1997). Nonmonotonic Reasoning. MIT Press (in press)

G. Brewka (1994). Reasoning about Priorities in Default Logic. In Proc. 12thNational Conference on Artificial Intelligence, AAAI/MIT Press, 940-945

D. Etherington (1987). Formalizing Nonmonotonic Reasoning Systems. ArtificialIntelligence 31: 41-85

W. Lukaszewicz (1988). Considerations on Default Logic.Computational Intelligence 4: 1-16

D. Makinson (1989). General theory of cumulative inference. In Proc. 2ndInternational Workshop on Non-Monotonic Reasoning, Springer LNAI 346

Page 62: Default Reasoning and Theory Changecs4415/2010/resources/gregoris.pdf · Now I have to take back my previous conclusion: nonmonotonicity . MP2- 4 Incomplete information Why not use

MP2-62

References (continued)

W. Marek & M. Truszczynski (1993). Nonmonotonic Logic. Springer

R. Reiter (1980). A Logic for Default Reasoning. Artificial Intelligence 13: 81-132

T. Schaub (1992). On Constrained Default Theories. In Proc. 10th EuropeanConference on Artificial Intelligence