Confluence Operators Negotiation as Pointwise Merging S ebastien - - PowerPoint PPT Presentation

confluence operators negotiation as pointwise merging
SMART_READER_LITE
LIVE PREVIEW

Confluence Operators Negotiation as Pointwise Merging S ebastien - - PowerPoint PPT Presentation

Confluence Operators Negotiation as Pointwise Merging S ebastien Konieczny Ram on Pino P erez CNRS - Centre de Recherche en Informatique de Lens (CRIL) Universit e dArtois, Lens, France konieczny@cril.fr Departamento de


slide-1
SLIDE 1

Confluence Operators – Negotiation as Pointwise Merging –

S´ ebastien Konieczny Ram´

  • n Pino P´

erez

CNRS - Centre de Recherche en Informatique de Lens (CRIL) Universit´ e d’Artois, Lens, France konieczny@cril.fr Departamento de Matem´ aticas, Facultad de ciencias Universidad de Los Andes, M´ erida, Venezuela pino@ula.ve

1 / 24

slide-2
SLIDE 2

Belief change operators

Revision Belief revision is the process of accomodating a new piece of evidence that is more reliable than the current beliefs of the agent. In belief revision the world is static, it is the beliefs of the agents that evolve. Update In belief update the new piece of evidence denotes a change in the

  • world. The world is dynamic, and these (observed) changes modify

the beliefs of the agent. Merging Belief merging is the process of defining the beliefs of a group of

  • agents. So the question is: Given a set of agents that have their
  • wn beliefs, what can be considered as the beliefs of the group?

2 / 24

slide-3
SLIDE 3

Motivation

Revision Merging Update

3 / 24

slide-4
SLIDE 4

Motivation

Revision Merging Update

3 / 24

slide-5
SLIDE 5

Motivation

Revision Merging Update

3 / 24

slide-6
SLIDE 6

Motivation

Revision Merging Update ?

3 / 24

slide-7
SLIDE 7

Motivation

Revision Merging Update Confluence

3 / 24

slide-8
SLIDE 8

Framework

  • Propositional logic:

A formula ϕ is build from

◮ A set P of propositional symbols (a, b, . . .) ◮ And logical connectives (¬, ∧, ∨, →,. . .)

An interpretation ω is a function from P to {0, 1} mod(ϕ) = {ω ∈ W | ω | = ϕ} A formula is complete if it has a unique model

  • A base ϕ is a (finite set of) propositional formula
  • A profile Ψ is a multi-set of bases : Ψ = {ϕ1, . . . , ϕn}

4 / 24

slide-9
SLIDE 9

Revision

Revision Belief revision is the process of accomodating a new piece of evidence that is more reliable than the current beliefs of the agent. In belief revision the world is static, it is the beliefs of the agents that evolve.

5 / 24

slide-10
SLIDE 10

Revision

Revision Belief revision is the process of accomodating a new piece of evidence that is more reliable than the current beliefs of the agent. In belief revision the world is static, it is the beliefs of the agents that evolve. 3 principles:

  • Primacy of update
  • Coherence
  • Minimal change

5 / 24

slide-11
SLIDE 11

Revision

Revision Belief revision is the process of accomodating a new piece of evidence that is more reliable than the current beliefs of the agent. In belief revision the world is static, it is the beliefs of the agents that evolve. 3 principles:

  • Primacy of update
  • Coherence
  • Minimal change

5 / 24

slide-12
SLIDE 12

Revision

Revision Belief revision is the process of accomodating a new piece of evidence that is more reliable than the current beliefs of the agent. In belief revision the world is static, it is the beliefs of the agents that evolve. 3 principles:

  • Primacy of update
  • Coherence
  • Minimal change

5 / 24

slide-13
SLIDE 13

Revision

Revision Belief revision is the process of accomodating a new piece of evidence that is more reliable than the current beliefs of the agent. In belief revision the world is static, it is the beliefs of the agents that evolve. 3 principles:

  • Primacy of update
  • Coherence
  • Minimal change

5 / 24

slide-14
SLIDE 14

Revision

Revision Belief revision is the process of accomodating a new piece of evidence that is more reliable than the current beliefs of the agent. In belief revision the world is static, it is the beliefs of the agents that evolve. 3 principles:

  • Primacy of update
  • Coherence
  • Minimal change

5 / 24

slide-15
SLIDE 15

Revision

ϕ = (b ∧ ¬m) ∨ (¬b ∧ m) µ = b

6 / 24

slide-16
SLIDE 16

Revision

ϕ = (b ∧ ¬m) ∨ (¬b ∧ m) mod(ϕ) = {10, 01} µ = b mod(µ) = {10, 11}

6 / 24

slide-17
SLIDE 17

Revision

ϕ = (b ∧ ¬m) ∨ (¬b ∧ m) mod(ϕ) = {10, 01} µ = b mod(µ) = {10, 11} ϕ ◦ µ = b ∧ ¬m

6 / 24

slide-18
SLIDE 18

Update

Update In belief update the new piece of evidence denotes a change in the

  • world. The world is dynamic, and these (observed) changes modify

the beliefs of the agent.

7 / 24

slide-19
SLIDE 19

Update

Update In belief update the new piece of evidence denotes a change in the

  • world. The world is dynamic, and these (observed) changes modify

the beliefs of the agent.

7 / 24

slide-20
SLIDE 20

Update

Update In belief update the new piece of evidence denotes a change in the

  • world. The world is dynamic, and these (observed) changes modify

the beliefs of the agent.

7 / 24

slide-21
SLIDE 21

Update

Update In belief update the new piece of evidence denotes a change in the

  • world. The world is dynamic, and these (observed) changes modify

the beliefs of the agent.

7 / 24

slide-22
SLIDE 22

Update

Update In belief update the new piece of evidence denotes a change in the

  • world. The world is dynamic, and these (observed) changes modify

the beliefs of the agent.

7 / 24

slide-23
SLIDE 23

Update

Update In belief update the new piece of evidence denotes a change in the

  • world. The world is dynamic, and these (observed) changes modify

the beliefs of the agent.

7 / 24

slide-24
SLIDE 24

Update

Update In belief update the new piece of evidence denotes a change in the

  • world. The world is dynamic, and these (observed) changes modify

the beliefs of the agent.

7 / 24

slide-25
SLIDE 25

Update

Update In belief update the new piece of evidence denotes a change in the

  • world. The world is dynamic, and these (observed) changes modify

the beliefs of the agent.

7 / 24

slide-26
SLIDE 26

Update

Update In belief update the new piece of evidence denotes a change in the

  • world. The world is dynamic, and these (observed) changes modify

the beliefs of the agent.

7 / 24

slide-27
SLIDE 27

Update

Update In belief update the new piece of evidence denotes a change in the

  • world. The world is dynamic, and these (observed) changes modify

the beliefs of the agent.

7 / 24

slide-28
SLIDE 28

Update

ϕ = (b ∧ ¬m) ∨ (¬b ∧ m) mod(ϕ) = {10, 01} µ = b mod(µ) = {10, 11}

8 / 24

slide-29
SLIDE 29

Update

ϕ = (b ∧ ¬m) ∨ (¬b ∧ m) mod(ϕ) = {10, 01} µ = b mod(µ) = {10, 11} ϕ ⋄ µ = (b ∧ ¬m)

8 / 24

slide-30
SLIDE 30

Update

ϕ = (b ∧ ¬m) ∨ (¬b ∧ m) mod(ϕ) = {10, 01} µ = b mod(µ) = {10, 11} ϕ ⋄ µ = (b ∧ ¬m)

8 / 24

slide-31
SLIDE 31

Update

ϕ = (b ∧ ¬m) ∨ (¬b ∧ m) mod(ϕ) = {10, 01} µ = b mod(µ) = {10, 11} ϕ ⋄ µ = (b ∧ ¬m) ∨ (b ∧ m)

8 / 24

slide-32
SLIDE 32

Update

ϕ = (b ∧ ¬m) ∨ (¬b ∧ m) mod(ϕ) = {10, 01} µ = b mod(µ) = {10, 11} ϕ ⋄ µ = (b ∧ ¬m) ∨ (b ∧ m)

8 / 24

slide-33
SLIDE 33

Merging

Merging Belief merging is the process of defining the beliefs of a group of

  • agents. So the question is: Given a set of agents that have their
  • wn beliefs, what can be considered as the beliefs of the group?

9 / 24

slide-34
SLIDE 34

Merging

Merging Belief merging is the process of defining the beliefs of a group of

  • agents. So the question is: Given a set of agents that have their
  • wn beliefs, what can be considered as the beliefs of the group?

ϕ1 ϕ2 ϕ3 a, b → c a, b ¬ a △({ϕ1, ϕ2, ϕ3}) =

9 / 24

slide-35
SLIDE 35

Merging

Merging Belief merging is the process of defining the beliefs of a group of

  • agents. So the question is: Given a set of agents that have their
  • wn beliefs, what can be considered as the beliefs of the group?

ϕ1 ϕ2 ϕ3 a, b → c a, b ¬ a △({ϕ1, ϕ2, ϕ3}) = b → c, b

9 / 24

slide-36
SLIDE 36

Merging

Merging Belief merging is the process of defining the beliefs of a group of

  • agents. So the question is: Given a set of agents that have their
  • wn beliefs, what can be considered as the beliefs of the group?

ϕ1 ϕ2 ϕ3 a, b → c a, b ¬ a △({ϕ1, ϕ2, ϕ3}) = b → c, b, a

9 / 24

slide-37
SLIDE 37

Confluence

g: german car e: expensive car s: sport car ϕ1 = ¬g ∧ ¬e ∧ s mod(ϕ1) = {001} ϕ2 = (g ∧ e ∧ s) ∨ (¬g ∧ ¬e ∧ s) mod(ϕ2) = {001, 111} µ = ¬(g ∧ ¬e ∧ s) mod(µ) = W \ {101}

  • Belief/Goal Merging:

△µ({ϕ1, ϕ2}) = ¬g ∧ ¬e ∧ s

10 / 24

slide-38
SLIDE 38

Confluence

g: german car e: expensive car s: sport car ϕ1 = ¬g ∧ ¬e ∧ s mod(ϕ1) = {001} ϕ2 = (g ∧ e ∧ s) ∨ (¬g ∧ ¬e ∧ s) mod(ϕ2) = {001, 111} µ = ¬(g ∧ ¬e ∧ s) mod(µ) = W \ {101}

  • Belief/Goal Merging:

△µ({ϕ1, ϕ2}) = ¬g ∧ ¬e ∧ s

10 / 24

slide-39
SLIDE 39

Confluence

g: german car e: expensive car s: sport car ϕ1 = ¬g ∧ ¬e ∧ s mod(ϕ1) = {001} ϕ2 = (g ∧ e ∧ s) ∨ (¬g ∧ ¬e ∧ s) mod(ϕ2) = {001, 111} µ = ¬(g ∧ ¬e ∧ s) mod(µ) = W \ {101}

  • Belief/Goal Merging:

△µ({ϕ1, ϕ2}) = ¬g ∧ ¬e ∧ s

  • Confluence:

♦µ({ϕ1, ϕ2}) = (¬g ∧ ¬e ∧ s) ∨ (¬g ∧ e ∧ s) = ¬g ∧ s 001 011

10 / 24

slide-40
SLIDE 40

Confluence

g: german car e: expensive car s: sport car ϕ1 = ¬g ∧ ¬e ∧ s mod(ϕ1) = {001} ϕ2 = (g ∧ e ∧ s) ∨ (¬g ∧ ¬e ∧ s) mod(ϕ2) = {001, 111} µ = ¬(g ∧ ¬e ∧ s) mod(µ) = W \ {101}

  • Belief/Goal Merging:

△µ({ϕ1, ϕ2}) = ¬g ∧ ¬e ∧ s

  • Confluence:

♦µ({ϕ1, ϕ2}) = (¬g ∧ ¬e ∧ s) ∨ (¬g ∧ e ∧ s) = ¬g ∧ s 001 011

10 / 24

slide-41
SLIDE 41

Confluence

g: german car e: expensive car s: sport car ϕ1 = ¬g ∧ ¬e ∧ s mod(ϕ1) = {001} ϕ2 = (g ∧ e ∧ s) ∨ (¬g ∧ ¬e ∧ s) mod(ϕ2) = {001, 111} µ = ¬(g ∧ ¬e ∧ s) mod(µ) = W \ {101}

  • Belief/Goal Merging:

△µ({ϕ1, ϕ2}) = ¬g ∧ ¬e ∧ s

  • Confluence:

♦µ({ϕ1, ϕ2}) = (¬g ∧ ¬e ∧ s) ∨ (¬g ∧ e ∧ s) = ¬g ∧ s 001 011

Unperfectly known goals Potential evolution

10 / 24

slide-42
SLIDE 42

Revision [Alchourr´

  • n-G¨

ardenfors-Makinson 85]

(R1) ϕ ◦ µ ⊢ µ (R2) If ϕ ∧ µ ⊥ then ϕ ◦ µ ≡ ϕ ∧ µ (R3) If µ ⊥ then ϕ ◦ µ ⊥ (R4) If ϕ1 ≡ ϕ2 and µ1 ≡ µ2 then ϕ1 ◦ µ1 ≡ ϕ2 ◦ µ2 (R5) (ϕ ◦ µ) ∧ φ ⊢ ϕ ◦ (µ ∧ φ) (R6) If (ϕ ◦ µ) ∧ φ ⊥ then ϕ ◦ (µ ∧ φ) ⊢ (ϕ ◦ µ) ∧ φ

11 / 24

slide-43
SLIDE 43

Revision [Alchourr´

  • n-G¨

ardenfors-Makinson 85]

(R1) ϕ ◦ µ ⊢ µ (R2) If ϕ ∧ µ ⊥ then ϕ ◦ µ ≡ ϕ ∧ µ (R3) If µ ⊥ then ϕ ◦ µ ⊥ (R4) If ϕ1 ≡ ϕ2 and µ1 ≡ µ2 then ϕ1 ◦ µ1 ≡ ϕ2 ◦ µ2 (R5) (ϕ ◦ µ) ∧ φ ⊢ ϕ ◦ (µ ∧ φ) (R6) If (ϕ ◦ µ) ∧ φ ⊥ then ϕ ◦ (µ ∧ φ) ⊢ (ϕ ◦ µ) ∧ φ A faithful assignment is a function mapping each base ϕ to a pre-order ≤ϕ over interpretations such that:

  • If ω |

= ϕ and ω′ | = ϕ, then ω ≃ϕ ω′

  • If ω |

= ϕ and ω′ | = ϕ, then ω <ϕ ω′

  • If ϕ ≡ ϕ′, then ≤ϕ=≤ϕ′

11 / 24

slide-44
SLIDE 44

Revision [Alchourr´

  • n-G¨

ardenfors-Makinson 85]

(R1) ϕ ◦ µ ⊢ µ (R2) If ϕ ∧ µ ⊥ then ϕ ◦ µ ≡ ϕ ∧ µ (R3) If µ ⊥ then ϕ ◦ µ ⊥ (R4) If ϕ1 ≡ ϕ2 and µ1 ≡ µ2 then ϕ1 ◦ µ1 ≡ ϕ2 ◦ µ2 (R5) (ϕ ◦ µ) ∧ φ ⊢ ϕ ◦ (µ ∧ φ) (R6) If (ϕ ◦ µ) ∧ φ ⊥ then ϕ ◦ (µ ∧ φ) ⊢ (ϕ ◦ µ) ∧ φ A faithful assignment is a function mapping each base ϕ to a pre-order ≤ϕ over interpretations such that:

  • If ω |

= ϕ and ω′ | = ϕ, then ω ≃ϕ ω′

  • If ω |

= ϕ and ω′ | = ϕ, then ω <ϕ ω′

  • If ϕ ≡ ϕ′, then ≤ϕ=≤ϕ′

Theorem (Katsuno-Mendelzon 91a)

An operator ◦ is a revision operator (ie. it satisfies (R1)-(R6)) if and only if there exists a faithful assignment that maps each base ϕ to a total pre-order ≤ϕ such that mod(ϕ ◦ µ) = min(mod(µ), ≤ϕ).

11 / 24

slide-45
SLIDE 45

Update [Kastuno-Mendelzon 91b]

(U1) ϕ ⋄ µ ⊢ µ (U2) If ϕ ⊢ µ, then ϕ ⋄ µ ≡ ϕ (U3) If ϕ ⊥ and µ ⊥ then ϕ ⋄ µ ⊥ (U4) If ϕ1 ≡ ϕ2 and µ1 ≡ µ2 then ϕ1 ⋄ µ1 ≡ ϕ2 ⋄ µ2 (U5) (ϕ ⋄ µ) ∧ φ ⊢ ϕ ⋄ (µ ∧ φ) (U6) If ϕ ⋄ µ1 ⊢ µ2 and ϕ ⋄ µ2 ⊢ µ1, then ϕ ⋄ µ1 ≡ ϕ ⋄ µ2 (U7) If ϕ is a complete formula, then (ϕ ⋄ µ1) ∧ (ϕ ⋄ µ2) ⊢ ϕ ⋄ (µ1 ∨ µ2) (U8) (ϕ1 ∨ ϕ2) ⋄ µ ≡ (ϕ1 ⋄ µ) ∨ (ϕ2 ⋄ µ) (U9) If ϕ is a complete formula and (ϕ ⋄ µ) ∧ φ ⊥, then ϕ ⋄ (µ ∧ φ) ⊢ (ϕ ⋄ µ) ∧ φ

Theorem

An update operator ⋄ satisfies (U1)-(U8) if and only if there exists a faithful assignment that maps each interpretation ω to a partial pre-order ≤ω such that mod(ϕ ⋄ µ) =

  • ω|

min(mod(µ), ≤ω)

12 / 24

slide-46
SLIDE 46

Update [Kastuno-Mendelzon 91b]

(U1) ϕ ⋄ µ ⊢ µ (U2) If ϕ ⊢ µ, then ϕ ⋄ µ ≡ ϕ (U3) If ϕ ⊥ and µ ⊥ then ϕ ⋄ µ ⊥ (U4) If ϕ1 ≡ ϕ2 and µ1 ≡ µ2 then ϕ1 ⋄ µ1 ≡ ϕ2 ⋄ µ2 (U5) (ϕ ⋄ µ) ∧ φ ⊢ ϕ ⋄ (µ ∧ φ) (U6) If ϕ ⋄ µ1 ⊢ µ2 and ϕ ⋄ µ2 ⊢ µ1, then ϕ ⋄ µ1 ≡ ϕ ⋄ µ2 (U7) If ϕ is a complete formula, then (ϕ ⋄ µ1) ∧ (ϕ ⋄ µ2) ⊢ ϕ ⋄ (µ1 ∨ µ2) (U8) (ϕ1 ∨ ϕ2) ⋄ µ ≡ (ϕ1 ⋄ µ) ∨ (ϕ2 ⋄ µ) (U9) If ϕ is a complete formula and (ϕ ⋄ µ) ∧ φ ⊥, then ϕ ⋄ (µ ∧ φ) ⊢ (ϕ ⋄ µ) ∧ φ

Theorem

An update operator ⋄ satisfies (U1)-(U5), (U8) and (U9) if and only if there exists a faithful assignment that maps each interpretation ω to a total pre-order ≤ω such that mod(ϕ ⋄ µ) =

  • ω|

min(mod(µ), ≤ω)

12 / 24

slide-47
SLIDE 47

Merging [Konieczny-Pino-P´ erez 99]

(IC0) △µ(Ψ) ⊢ µ (IC1) If µ is consistent, then △µ(Ψ) is consistent (IC2) If Ψ is consistent with µ, then △µ(Ψ) ≡ Ψ ∧ µ (IC3) If Ψ1 ≡ Ψ2 and µ1 ≡ µ2, then △µ1(Ψ1) ≡ △µ2(Ψ2) (IC4) If ϕ1 ⊢ µ and ϕ2 ⊢ µ, then △µ({ϕ1, ϕ2}) ∧ ϕ1 is consistent if and

  • nly if △µ({ϕ1, ϕ2}) ∧ ϕ2 is consistent

(IC5) △µ(Ψ1) ∧ △µ(Ψ2) ⊢ △µ(Ψ1 ⊔ Ψ2) (IC6) If △µ(Ψ1) ∧ △µ(Ψ2) is consistent, then △µ(Ψ1 ⊔ Ψ2) ⊢ △µ(Ψ1) ∧ △µ(Ψ2) (IC7) △µ1(Ψ) ∧ µ2 ⊢ △µ1∧µ2(Ψ) (IC8) If △µ1(Ψ) ∧ µ2 is consistent, then △µ1∧µ2(Ψ) ⊢ △µ1(Ψ)

13 / 24

slide-48
SLIDE 48

Merging [Konieczny-Pino-P´ erez 99]

A syncretic assignment is a function mapping each profile Ψ to a total pre-order ≤Ψ over interpretations such that:

  • If ω |

= Ψ and ω′ | = Ψ, then ω ≃Ψ ω′

  • If ω |

= Ψ and ω′ | = Ψ, then ω <Ψ ω′

  • If Ψ1 ≡ Ψ2, then ≤Ψ1=≤Ψ2
  • ∀ω |

= ϕ ∃ω′ | = ϕ′ ω′ ≤{ϕ}⊔{ϕ′} ω

  • If ω ≤Ψ1 ω′ and ω ≤Ψ2 ω′, then ω ≤Ψ1⊔Ψ2 ω′
  • If ω <Ψ1 ω′ and ω ≤Ψ2 ω′, then ω <Ψ1⊔Ψ2 ω′

Theorem

An operator △ is an IC merging operator if and only if there exists a syncretic assignment that maps each profile Ψ to a total pre-order ≤Ψ such that mod(△µ(Ψ)) = min(mod(µ), ≤Ψ)

14 / 24

slide-49
SLIDE 49

Revision vs Update

Proposition

If ◦ is a revision operator (i.e. it satisfies (R1)-(R6)), then the operator ⋄ defined by: ϕ ⋄ µ =

  • ω|

ϕω ◦ µ is an update operator that satisfies (U1)-(U9). Moreover, for each update operator ⋄, there exists a revision operator ◦ such that the previous equation holds.

15 / 24

slide-50
SLIDE 50

Revision vs Merging

Proposition

If △ is an IC merging operator (it satisfies (IC0-IC8)), then the operator ◦, defined as ϕ ◦ µ = △µ(ϕ), is an AGM revision operator (it satisfies (R1-R6)).

16 / 24

slide-51
SLIDE 51

Confluence

Revision Merging Update

slide-52
SLIDE 52

Confluence

Revision Merging Update ?

17 / 24

slide-53
SLIDE 53

Confluence

An operator ♦ is a confluence operator if it satisfies the following properties: (UC0) ♦µ(Ψ) ⊢ µ (UC1) If µ is consistent and Ψ is p-consistent, then ♦µ(Ψ) is consistent (UC2) If Ψ is complete, Ψ is consistent and Ψ ⊢ µ, then ♦µ(Ψ) ≡ Ψ (UC3) If Ψ1 ≡ Ψ2 and µ1 ≡ µ2, then ♦µ1(Ψ1) ≡ ♦µ2(Ψ2) (UC4) If ϕ1 and ϕ2 are complete formulae and ϕ1 ⊢ µ, ϕ2 ⊢ µ, then ♦µ({ϕ1, ϕ2}) ∧ ϕ1 is consistent if and only ♦µ({ϕ1, ϕ2}) ∧ ϕ2 is consistent (UC5) ♦µ(Ψ1) ∧ ♦µ(Ψ2) ⊢ ♦µ(Ψ1 ⊔ Ψ2) (UC6) If Ψ1 and Ψ2 are complete profiles and ♦µ(Ψ1) ∧ ♦µ(Ψ2) is consistent, then ♦µ(Ψ1 ⊔ Ψ2) ⊢ ♦µ(Ψ1) ∧ ♦µ(Ψ2) (UC7) ♦µ1(Ψ) ∧ µ2 ⊢ ♦µ1∧µ2(Ψ) (UC8) If Ψ is a complete profile and if ♦µ1(Ψ) ∧ µ2 is consistent then ♦µ1∧µ2(Ψ) ⊢ ♦µ1(Ψ) ∧ µ2 (UC9) ♦µ(Ψ ⊔ {ϕ ∨ ϕ′}) ≡ ♦µ(Ψ ⊔ {ϕ}) ∨ ♦µ(Ψ ⊔ {ϕ′})

18 / 24

slide-54
SLIDE 54

State

Definition

  • A multi-set of interpretations e will be called a state.
  • If Ψ = {ϕ1, . . . , ϕn} is a profile and e = {ω1, . . . , ωn} is a state such that

ωi | = ϕi for each i, we say that e is a state of the profile Ψ, that will be denoted by e | = Ψ.

  • If e = {ω1, . . . , ωn} is a state, we define the profile Ψe by putting

Ψe = {ϕ{ω1}, . . . , ϕ{ωn}}

Lemma

If ♦ satisfies (UC3) and (UC9) then ♦ satisfies the following ♦µ(Ψ) ≡

  • e|

♦µ(Ψe)

19 / 24

slide-55
SLIDE 55

Representation theorem

A distributed assignment is a function mapping each state e to a total pre-order ≤e over interpretations such that:

  • ω <{ω,...,ω} ω′ if ω′ = ω
  • ω ≃{ω,ω′} ω′
  • If ω ≤e1 ω′ and ω ≤e2 ω′, then ω ≤e1⊔e2 ω′
  • If ω <e1 ω′ and ω ≤e2 ω′, then ω <e1⊔e2 ω′

Theorem

An operator ♦ is a confluence operator if and only if there exists a distributed assignment that maps each state e to a total pre-order ≤e such that mod(♦µ(Ψ)) =

  • e|

min(mod(µ), ≤e) (1)

20 / 24

slide-56
SLIDE 56

Confluence vs Update and Merging

Proposition

If ♦ is a confluence operator (i.e. it satisfies (UC0-UC9)), then the operator ⋄, defined as ϕ ⋄ µ = ♦µ(ϕ), is an update operator (i.e. it satisfies (U1-U9)).

Proposition

If △ is an IC merging operator (i.e. it satisfies (IC0-IC8)) then the operator ♦ defined by ♦µ(Ψ) =

  • e|

△µ(Ψe) is a confluence operator (i.e. it satisfies (UC0-UC9)). Moreover, for each confluence operator ♦, there exists a merging operator △ such that the previous equation holds.

21 / 24

slide-57
SLIDE 57

Example of confluence operators

  • A distance d between interpretations

Drastic distance, Hamming (Dalal) distance, . . .

  • An aggregation function f

sum, leximax, . . .

  • ω ≤e ω′ if and only if d(ω, e) ≤ d(ω′, e), where (e = {ω1, . . . , ωn}):

d(ω, e) = f (d(ω, ω1) . . . , d(ω, ωn))

  • mod(♦µ(Ψ)) =

e| =Ψ min(mod(µ), ≤e)

22 / 24

slide-58
SLIDE 58

Example

Let Ψ = {ϕ1, ϕ2} and µ : mod(µ) = W \ {101} mod(ϕ1) = {001} mod(ϕ2) = {001, 111} The corresponding states are: e1 = {001, 001} e2 = {001, 111}

W 001 111 e1 e2 ♦dH,Σ

µ

♦dH,Gmax

µ

Σ

Gmax

Σ

Gmax

000 1 3 2 11 4 31 001 2 00 2 20 × × 010 2 2 4 22 4 22 011 1 1 2 11 2 11 × × 100 2 2 4 22 4 22 101 1 1 2 11 2 11 110 3 1 6 33 4 31 111 2 4 22 2 20 ×

23 / 24

slide-59
SLIDE 59

Example

Let Ψ = {ϕ1, ϕ2} and µ : mod(µ) = W \ {101} mod(ϕ1) = {001} mod(ϕ2) = {001, 111} The corresponding states are: e1 = {001, 001} e2 = {001, 111}

W 001 111 e1 e2 ♦dH,Σ

µ

♦dH,Gmax

µ

Σ

Gmax

Σ

Gmax

000 1 3 2 11 4 31 001 2 00 2 20 × × 010 2 2 4 22 4 22 011 1 1 2 11 2 11 × × 100 2 2 4 22 4 22 101 1 1 2 11 2 11 110 3 1 6 33 4 31 111 2 4 22 2 20 ×

mod(♦dH,Σ

µ

(Ψ)) = {001, 011, 111}

23 / 24

slide-60
SLIDE 60

Example

Let Ψ = {ϕ1, ϕ2} and µ : mod(µ) = W \ {101} mod(ϕ1) = {001} mod(ϕ2) = {001, 111} The corresponding states are: e1 = {001, 001} e2 = {001, 111}

W 001 111 e1 e2 ♦dH,Σ

µ

♦dH,Gmax

µ

Σ

Gmax

Σ

Gmax

000 1 3 2 11 4 31 001 2 00 2 20 × × 010 2 2 4 22 4 22 011 1 1 2 11 2 11 × × 100 2 2 4 22 4 22 101 1 1 2 11 2 11 110 3 1 6 33 4 31 111 2 4 22 2 20 ×

mod(♦dH,Σ

µ

(Ψ)) = {001, 011, 111} mod(♦dH,Gmax

µ

(Ψ)) = {001, 011}

23 / 24

slide-61
SLIDE 61

Conclusion

  • Confluence operators
  • Pointwise merging
  • Negotiation
  • Belief vs Goal aggregation

24 / 24