strategies for distributed cbr santi ontañón iiia-csic
TRANSCRIPT
![Page 1: Strategies for Distributed CBR Santi Ontañón IIIA-CSIC](https://reader036.vdocuments.mx/reader036/viewer/2022062322/5697bf781a28abf838c81b9e/html5/thumbnails/1.jpg)
Strategies for Distributed CBR
Santi Ontañón
IIIA-CSIC
![Page 2: Strategies for Distributed CBR Santi Ontañón IIIA-CSIC](https://reader036.vdocuments.mx/reader036/viewer/2022062322/5697bf781a28abf838c81b9e/html5/thumbnails/2.jpg)
OUTLINE
CBR Introduction Applications
Distributed CBR Autonomous CBR agents Collaboration policies Learning to collaborate Case Bartering
Conclusions
![Page 3: Strategies for Distributed CBR Santi Ontañón IIIA-CSIC](https://reader036.vdocuments.mx/reader036/viewer/2022062322/5697bf781a28abf838c81b9e/html5/thumbnails/3.jpg)
What is CBR?
Case Based Reasoning: To solve a new
problem by noticing its similarity to one or
several previously solved problems and
by adapting their solutions instead of
working out a solution from scratch.
![Page 4: Strategies for Distributed CBR Santi Ontañón IIIA-CSIC](https://reader036.vdocuments.mx/reader036/viewer/2022062322/5697bf781a28abf838c81b9e/html5/thumbnails/4.jpg)
GeneralKnowledge
CBR: Cycle
CASEBASE Retrieved
cases
Newcase
TestedCase
Solvedcase
RETRIEVE
REUSEREVISE
RETAIN
PROBLEM
![Page 5: Strategies for Distributed CBR Santi Ontañón IIIA-CSIC](https://reader036.vdocuments.mx/reader036/viewer/2022062322/5697bf781a28abf838c81b9e/html5/thumbnails/5.jpg)
CBR: Retrieve
Is the key problem in CBR
Indexing:
Relevant features: predictive, discriminatory
and explanatory.
Good case representation
Similarity metrics
![Page 6: Strategies for Distributed CBR Santi Ontañón IIIA-CSIC](https://reader036.vdocuments.mx/reader036/viewer/2022062322/5697bf781a28abf838c81b9e/html5/thumbnails/6.jpg)
CBR: Reuse
Copy solution (for classification)
Modify old solution: Reinstantiate (variables)
Local search
Transform Domain knowledge (causal model)
![Page 7: Strategies for Distributed CBR Santi Ontañón IIIA-CSIC](https://reader036.vdocuments.mx/reader036/viewer/2022062322/5697bf781a28abf838c81b9e/html5/thumbnails/7.jpg)
CBR: Revise
Apply solution to problem: In real world In simulated world
Evaluate Repair
User guided Internal (Domain knowledge)
![Page 8: Strategies for Distributed CBR Santi Ontañón IIIA-CSIC](https://reader036.vdocuments.mx/reader036/viewer/2022062322/5697bf781a28abf838c81b9e/html5/thumbnails/8.jpg)
CBR: Retain
What to retain? Relevant problems
Solution (failed/successful)
Solution method
Store case Update/indentify indexes
![Page 9: Strategies for Distributed CBR Santi Ontañón IIIA-CSIC](https://reader036.vdocuments.mx/reader036/viewer/2022062322/5697bf781a28abf838c81b9e/html5/thumbnails/9.jpg)
CBR systems:
CHEF, cooking planing (Hammon 86) HYPO, legal reasoning (Ashley/Rissland
87) PROTOS, medical diagnosis
(Bereiss/Porter 88) CASEY, medial diagnosis (Koton 89) SAXEX, expresive music (Arcos 96)
![Page 10: Strategies for Distributed CBR Santi Ontañón IIIA-CSIC](https://reader036.vdocuments.mx/reader036/viewer/2022062322/5697bf781a28abf838c81b9e/html5/thumbnails/10.jpg)
Distributed CBR
Autonomous CBR agents
Collaboration policies
Learning to collaborate
Case Bartering
![Page 11: Strategies for Distributed CBR Santi Ontañón IIIA-CSIC](https://reader036.vdocuments.mx/reader036/viewer/2022062322/5697bf781a28abf838c81b9e/html5/thumbnails/11.jpg)
Autonomous CBR agents
Multiagent CBR system Each agent has an individual Case-Base
Agent has autonomy in Problem acquisition Problem solving Learning
Collaboration policies Explicit strategies to improve individual performance
![Page 12: Strategies for Distributed CBR Santi Ontañón IIIA-CSIC](https://reader036.vdocuments.mx/reader036/viewer/2022062322/5697bf781a28abf838c81b9e/html5/thumbnails/12.jpg)
Multiagent CBR system
MAC (Multiagent CBR) System
An agent is a pair (Ai,Ci)
A Case base
Classification task in
€
{(Pji,Sk)}j=1...Ni
€
{(Ai,Ci)}i=1...n
€
{S1,...,SK}
![Page 13: Strategies for Distributed CBR Santi Ontañón IIIA-CSIC](https://reader036.vdocuments.mx/reader036/viewer/2022062322/5697bf781a28abf838c81b9e/html5/thumbnails/13.jpg)
Collaboration policies
Committee
Peer Counsel
Bounded Counsel
![Page 14: Strategies for Distributed CBR Santi Ontañón IIIA-CSIC](https://reader036.vdocuments.mx/reader036/viewer/2022062322/5697bf781a28abf838c81b9e/html5/thumbnails/14.jpg)
Collaboration
An agent A1 with problem P
Sends P to some agents A={ A2,…,An }
Each Aj answers with a Solution Endorsing
Record (SER)
• < { (S1 Ej1)…(Sm Ej
m) } , P , Aj >
Endorsing pairs: ( Sk,Ejk )
![Page 15: Strategies for Distributed CBR Santi Ontañón IIIA-CSIC](https://reader036.vdocuments.mx/reader036/viewer/2022062322/5697bf781a28abf838c81b9e/html5/thumbnails/15.jpg)
Solution EndorsingRecord (SER)
P4
S3
P1
S1
P3
S2
P5
S2
P8
S1
P7
S3
P2
S1
P6
S2
P9
S4
{ (S2,1), (S3,2) }
P
Relevant cases
![Page 16: Strategies for Distributed CBR Santi Ontañón IIIA-CSIC](https://reader036.vdocuments.mx/reader036/viewer/2022062322/5697bf781a28abf838c81b9e/html5/thumbnails/16.jpg)
Voting(1)
At is the set of agents submitting SER at time t (including the initiating agent).
Each agent has 1 vote The vote can be fractionally assigned to
various solution classes:
€
Vote(Sk,Ai) =Ek
i
c+ Eki
r=1...K∑
![Page 17: Strategies for Distributed CBR Santi Ontañón IIIA-CSIC](https://reader036.vdocuments.mx/reader036/viewer/2022062322/5697bf781a28abf838c81b9e/html5/thumbnails/17.jpg)
Voting(2)
The agent that receives de SERs, aggregates the votes:
The proposed solution is that with maximum Ballot:
€
Ballott(Sk,At)= Vote(Sk,Ai)
Ai ∈At
∑
€
Solt(P,At) =argmaxk=1...K
Ballott(Sk,At)( )
![Page 18: Strategies for Distributed CBR Santi Ontañón IIIA-CSIC](https://reader036.vdocuments.mx/reader036/viewer/2022062322/5697bf781a28abf838c81b9e/html5/thumbnails/18.jpg)
Committee policy (1)
An agent A1 with problem P
Sends P to all the other agents A={ A2,…,An }
Each Aj answers with a Solution Endorsing
Record (SER)
• < { (S1 Ej1)…(Sm Ej
m) } , P , Aj >
The majority vote selects the solution class
![Page 19: Strategies for Distributed CBR Santi Ontañón IIIA-CSIC](https://reader036.vdocuments.mx/reader036/viewer/2022062322/5697bf781a28abf838c81b9e/html5/thumbnails/19.jpg)
Agent interaction: Committee
A1 C1
P
A3 C3
A2 C2
![Page 20: Strategies for Distributed CBR Santi Ontañón IIIA-CSIC](https://reader036.vdocuments.mx/reader036/viewer/2022062322/5697bf781a28abf838c81b9e/html5/thumbnails/20.jpg)
Agent interaction: Committee
A1 C1
P
A3 C3
A2 C2P
P
![Page 21: Strategies for Distributed CBR Santi Ontañón IIIA-CSIC](https://reader036.vdocuments.mx/reader036/viewer/2022062322/5697bf781a28abf838c81b9e/html5/thumbnails/21.jpg)
Agent interaction: Committee
A1 C1
P SER of A1
A3 C3
A2 C2SER of A2
SER of A3
![Page 22: Strategies for Distributed CBR Santi Ontañón IIIA-CSIC](https://reader036.vdocuments.mx/reader036/viewer/2022062322/5697bf781a28abf838c81b9e/html5/thumbnails/22.jpg)
Agent interaction: Committee
A1 C1
P SER of A1
SER of A2
SOLUTION
SER of A3
A3 C3
A2 C2
Voting
![Page 23: Strategies for Distributed CBR Santi Ontañón IIIA-CSIC](https://reader036.vdocuments.mx/reader036/viewer/2022062322/5697bf781a28abf838c81b9e/html5/thumbnails/23.jpg)
Committee policy (2)
Robust Good accuracy even with big number of
agents with small case-bases Cost
Linear with number of agents Always asks all agents
![Page 24: Strategies for Distributed CBR Santi Ontañón IIIA-CSIC](https://reader036.vdocuments.mx/reader036/viewer/2022062322/5697bf781a28abf838c81b9e/html5/thumbnails/24.jpg)
Peer Counsel
The agent try to solve the problem by itself Assess competence of current solution:
Votes for the max class >> rest of votes:
If not competent, ask all other agents Like committee
€
Vmaxt
Vrestt >η
![Page 25: Strategies for Distributed CBR Santi Ontañón IIIA-CSIC](https://reader036.vdocuments.mx/reader036/viewer/2022062322/5697bf781a28abf838c81b9e/html5/thumbnails/25.jpg)
Agent interaction:Peer Counsel
A1 C1
P
A3 C3
A2 C2
![Page 26: Strategies for Distributed CBR Santi Ontañón IIIA-CSIC](https://reader036.vdocuments.mx/reader036/viewer/2022062322/5697bf781a28abf838c81b9e/html5/thumbnails/26.jpg)
Agent interaction:Peer Counsel
A1 C1
P
A3 C3
A2 C2
SER of A1
Competent!
![Page 27: Strategies for Distributed CBR Santi Ontañón IIIA-CSIC](https://reader036.vdocuments.mx/reader036/viewer/2022062322/5697bf781a28abf838c81b9e/html5/thumbnails/27.jpg)
Agent interaction:Peer Counsel
A1 C1
P
A3 C3
A2 C2
SER of A1
SOLUTION
Voting
![Page 28: Strategies for Distributed CBR Santi Ontañón IIIA-CSIC](https://reader036.vdocuments.mx/reader036/viewer/2022062322/5697bf781a28abf838c81b9e/html5/thumbnails/28.jpg)
Bounded Counsel Use self-competence model If not competent, ask one agent
Agent returns a SER Assess competence of current solution (Termination
Check):
• Votes for the max class >> rest of votes:
If not competent, ask one agent more
€
Vmaxt
Vrestt >η
![Page 29: Strategies for Distributed CBR Santi Ontañón IIIA-CSIC](https://reader036.vdocuments.mx/reader036/viewer/2022062322/5697bf781a28abf838c81b9e/html5/thumbnails/29.jpg)
Agent interaction:Bounded Counsel
A1 C1
P
A3 C3
A2 C2
![Page 30: Strategies for Distributed CBR Santi Ontañón IIIA-CSIC](https://reader036.vdocuments.mx/reader036/viewer/2022062322/5697bf781a28abf838c81b9e/html5/thumbnails/30.jpg)
Agent interaction:Bounded Counsel
A1 C1
P SER of A1
A3 C3
A2 C2
Not competent!
![Page 31: Strategies for Distributed CBR Santi Ontañón IIIA-CSIC](https://reader036.vdocuments.mx/reader036/viewer/2022062322/5697bf781a28abf838c81b9e/html5/thumbnails/31.jpg)
Agent interaction:Bounded Counsel
A1 C1
P SER of A1
A3 C3
A2 C2P
![Page 32: Strategies for Distributed CBR Santi Ontañón IIIA-CSIC](https://reader036.vdocuments.mx/reader036/viewer/2022062322/5697bf781a28abf838c81b9e/html5/thumbnails/32.jpg)
Agent interaction:Bounded Counsel
A1 C1
P SER of A1
A3 C3
A2 C2SER of A2
![Page 33: Strategies for Distributed CBR Santi Ontañón IIIA-CSIC](https://reader036.vdocuments.mx/reader036/viewer/2022062322/5697bf781a28abf838c81b9e/html5/thumbnails/33.jpg)
Agent interaction:Bounded Counsel
A1 C1
P SER of A1
A3 C3
A2 C2
SER of A2
Competent!
![Page 34: Strategies for Distributed CBR Santi Ontañón IIIA-CSIC](https://reader036.vdocuments.mx/reader036/viewer/2022062322/5697bf781a28abf838c81b9e/html5/thumbnails/34.jpg)
Agent interaction:Bounded Counsel
A1 C1
P SER of A1
A3 C3
A2 C2
SER of A2
SOLUTION
Voting
![Page 35: Strategies for Distributed CBR Santi Ontañón IIIA-CSIC](https://reader036.vdocuments.mx/reader036/viewer/2022062322/5697bf781a28abf838c81b9e/html5/thumbnails/35.jpg)
Learning to collaborate
Bounded Counsel fixed predicate to assess competence
Proactive Learning Improve collaboration Actions performed in order to learn learning when to ask counsel
![Page 36: Strategies for Distributed CBR Santi Ontañón IIIA-CSIC](https://reader036.vdocuments.mx/reader036/viewer/2022062322/5697bf781a28abf838c81b9e/html5/thumbnails/36.jpg)
Learning to collaborate(2)
A1 C1
P SER of A1
SER of A2
SER of A3
Do I need thecounsel of
another agent?
CANDIDATESOLUTION
Voting
![Page 37: Strategies for Distributed CBR Santi Ontañón IIIA-CSIC](https://reader036.vdocuments.mx/reader036/viewer/2022062322/5697bf781a28abf838c81b9e/html5/thumbnails/37.jpg)
Case Bartering
Why? Biased individual case-bases
How? Estimating underlying distribution
Interchanging cases between agents
![Page 38: Strategies for Distributed CBR Santi Ontañón IIIA-CSIC](https://reader036.vdocuments.mx/reader036/viewer/2022062322/5697bf781a28abf838c81b9e/html5/thumbnails/38.jpg)
Case Bartering(2)
Estimation of underlying distribution: Aggregating statistics of all the individual
case-bases.
€
D j =dij
i=1
n∑dil
l=1
K∑i=1
n∑
€
D={D1,...,DK}
![Page 39: Strategies for Distributed CBR Santi Ontañón IIIA-CSIC](https://reader036.vdocuments.mx/reader036/viewer/2022062322/5697bf781a28abf838c81b9e/html5/thumbnails/39.jpg)
Case Bartering(3)
Case-Base bias:
The agents should minimize its bias
€
Bias(Ai) = Dl −dil
dij
j=1
K∑
⎛
⎝
⎜ ⎜ ⎜
⎞
⎠
⎟ ⎟ ⎟
2
l=1
K
∑
![Page 40: Strategies for Distributed CBR Santi Ontañón IIIA-CSIC](https://reader036.vdocuments.mx/reader036/viewer/2022062322/5697bf781a28abf838c81b9e/html5/thumbnails/40.jpg)
Case Bartering(4)
Bartering Protocol:
1. Estimate underlying distribution
2. Send offers
3. Confirm offers
4. Interchange cases
5. If changes go to 2, else END.
![Page 41: Strategies for Distributed CBR Santi Ontañón IIIA-CSIC](https://reader036.vdocuments.mx/reader036/viewer/2022062322/5697bf781a28abf838c81b9e/html5/thumbnails/41.jpg)
Conclusions
Multi-agent systems offer new paradigms to organize AI.
Agent interaction is still a stimulating challenge.
CBR fits perfectly with these kind of applications.