trabalhos de pesquisa andrew diniz da costa andrew@les.inf.puc-rio.br
Post on 17-Apr-2015
113 Views
Preview:
TRANSCRIPT
Trabalhos de Pesquisa
Andrew Diniz da Costa
andrew@les.inf.puc-rio.br
2 © LES/PUC-Rio
Trabalho Realizados
2004 2006 2008 2009
Framework ASF
Agent Society Framework
4 © LES/PUC-Rio
Introdução
• MAS-ML
• Framework para aplicar conceitos presentes na linguagem de modelagem
– Ex: Papel, organização, ambiente ativo, etc.
• Agent Society Framework – ASF
– Criado no LES.
– Framework orientado a objetos para implementar sociedades de agentes.
– Conceitos: agente, Papel, objetivo, organização, ambiente, etc.
5 © LES/PUC-Rio
Introdução
• Foundation for Intelligent Physical Agents - FIPA
– Organização dedicada a desenvolver especificações na área de SMA.
– FIPA compliant
• Comunicação de agentes (ACL)
• Gerenciamento de agentes
• Transporte de mensagens (MTS)
– Jade, Jadex, etc.
6 © LES/PUC-Rio
Contribuições ao ASF
• Tornar o framework FIPA compliant– Serviço de transporte de mensagens
• 1 mensagem para 1..N receptores– Mensagens no formato ACL (Agent Communication Language)– Criação do supervisor AMS
• Criação• Alteração• Exclusão• Buscas
– Criação de uma série de classes e interfaces• Descrições de entidades• Interfaces definido os estados de ciclos de vida
7 © LES/PUC-Rio
Contribuições ao ASF
• Distribuição de agentes
– Agentes na mesma ou em máquinas diferentes podem interagir.
– Uma mensagem pode ser entregue para receptores localizados em diferentes máquinas.
– Comunicação por socket.
– Possibilidade de criar outras formas de comunicação (hotspots).
8 © LES/PUC-Rio
Contribuições ao ASF
• Melhorias gerais
– Novo conceito de ambiente ativo;
– Entidades melhor representadas (ex: agente, organização, etc.). Novos atributos e funcionalidades;
– Documentação completa do framework e de instâncias que servem como exemplo.
– Site do ASF (http://www.les.inf.puc-rio.br/frameworkasf/).
Agentes
Organizações
Papéis
Ambiente
9 © LES/PUC-Rio
Diagrama de Classe
10 © LES/PUC-Rio
Comparações com frameworks SMA
• Comparações com três frameworks: Jade, Jadex e Jack.
• Diferencial do ASF
– Melhor representação do conceito sociedade de agente.
– Maior liberdade da instância em relação à restrições do framework (Ex: política de seleção de planos).
– FIPA compliant.
– As instâncias utilizam o framework sem restrições de orientação objeto (desenvolvimento em Java).
11 © LES/PUC-Rio
Uso do ASF
• Alunos de mestrado e doutorado
SILVA, V. T. ; DURAN, Feranda ; GUEDES, José ; LUCENA, Carlos José Pereira de . Governing Multi-Agent Systems. Journal of the Brazilian Computer Society, v. 2 (13), p. 19-34, 2007.
12 © LES/PUC-Rio
Trabalho Realizados
2004 2006 2008 2009
ART-TestbedMASSES
Agent Reputation Trust (ART) Testbed(The finalist ZeCariocaLes)
14 © LES/PUC-Rio
Motivation
• The reputation concept is a complex information.
– Someone is honest or liar.
– The information supplied by someone is good or bad.
• How can the reputation help us? What are the ways?
• Use ideas of a framework created by old students of master, José Guedes e Fernanda Duran.
• Participate of some competition.
15 © LES/PUC-Rio
Competition
• Agent Reputation Trust (ART) Testbed
• Competition with agents
• AAMAS Conference
• Domain: appraisals for paintings
• Clients request appraisals for paintings from different eras
16 © LES/PUC-Rio
Competition
Agent 1
era1 era2 era9... era10
Agent 2
era1 era2 era9... era10
ZeCariocaLes
era1 era2 era9... era101,0 0,1 0,5 0,7
painting era1*
17 © LES/PUC-Rio
Competition
• It is necessary to complete the knowledge of each agent
• So, transactions with other agents should be executed.
• There are two types of transaction:
– Opinion
– Reputation
18 © LES/PUC-Rio
Transactions between agents
19 © LES/PUC-Rio
Game
• Each game has a lot of sessions. There isn’t a fix number.
• When a session finishes:– The true value of the paintings is disclosed.
– It is verified what agent got the best appraisals.
• In the current session each agent has the following information of the previous session:– The true value of the paintings
– The value of each opinion supplied by other agents
– ...
• The winner is the agent that has more money in the end of the game (the best financial administrator)
20 © LES/PUC-Rio
Lasts tests before the competition
21 © LES/PUC-Rio
Lasts tests before the competition
22 © LES/PUC-Rio
Competition
• 17 agents (1 didn’t execute) of 13 different institutions
• Two phases– Preliminary– Final
• Preliminary phase (May 10-11)– 8 agents of the different institutions– 15 agents offered by competition (5 “bad”, 5 “neutral”, 5 “bad”
dummies )– 100 rounds
• Final phase (May 16-17)– 5 best agents of the preliminary phase– 15 agents offered by competition (5 “bad”, 5 “neutral”, 5 “bad”
dummies )– 200 rounds
23 © LES/PUC-Rio
Preliminary Phase
24 © LES/PUC-Rio
Final Phase
5) Pontifícia Universidade Católica do Rio de Janeiro
4) Agents Research Lab, University of Girona
3) Department of Computer Engineering, Bogazici University
2) Department of Math & Computer Science, The University of Tulsa
1) Electronics & Computer Science, University of Southampton
25 © LES/PUC-Rio
Trabalho Realizados
2004 2006 2008 2009
ART-TestbedMASSES
Multi-Agent System for Stock Exchange Simulation
27 © LES/PUC-Rio
MASSES
• Domínio mais real para aplicar agentes de software
• Domínio de bolsa de valores
• Agentes são investidores da bolsa
• Cada dia é uma rodada do jogo.
28 © LES/PUC-Rio
Idéia Geral
29 © LES/PUC-Rio
Site do MASSES
30 © LES/PUC-Rio
Trabalho Realizados
2004 2006 2008 2009
DRP-MAS
A Hybrid Diagnostic-Recommendation Approach for Multi-Agent Systems
32 © LES/PUC-Rio
Motivation
• Governance Framework
• Multi-agent systems are societies with autonomous and heterogeneous agents, which can work together to achieve similar or different goals.
• The reason for some agent not to achieve some goal.
• Buyer desires to buy some product from some seller.– If the goal was not achieved then which was the reason?
– What to do?
33 © LES/PUC-Rio
Motivation
• Reputation concept related with diagnoses and recommendation
• Ubiquitous Computing Systems provide several situations that need of diagnoses and recommendations
34 © LES/PUC-Rio © LES/PUC-Rio
Difficulties of Diagnosing and Providing Alternative Executions
• We analyzed a set of points that deserved our attention during the creation of the new module
1. Deciding how to analyze the execution of the agents
2. Selecting data for diagnosing
3. Determining strategies for diagnoses
4. Determining trustworthy agents
5. Determining strategies for recommendations
6. Representing profiles of agents
7. Different devices (cell phones, laptops, PDA)• Limitations of hardware
8. Types of connection • Speed of connection (56Kbps, 512Kbps, etc), IP.
35 © LES/PUC-Rio
MediatorAgent
RequesterAgent
RecommendationAgent
DiagnosticAgent
(2)<<create>>
(2)<<create>>(1)
Request name of theDiagnosis Agent
(5)Provide name of the
Diagnosis Agent
(3)Send the
Recommendation name
(4)Send the
Requester name
General Idea
36 © LES/PUC-Rio
General Idea
RequesterAgent
DiagnosticAgent
RecommendationAgent
(1)
Request advices / Supply inform
ation, such as, quality of service
(2)Provide diagnosis
result
(3)Provide advices
Plan database
37 © LES/PUC-Rio
Solicitador A
Solicitador B
Mediador A
Mediador B
AgenteDiagnóstico A
Tipo de Diagnóstico 1
AgenteDiagnóstico B
Tipo de Diagnóstico 2
AgenteRecomendação A
AgenteRecomendação B
Tipo de Recomendação
Requisita
Provê
Requisita
Provê
<<criar>>
<<criar>>
<<criar>><<criar>>
General Idea
38 © LES/PUC-Rio
Architecture
Application
Mediation
Diagnosis
Recommendation
Artificial IntelligenceToolset
DRP-MAS
Reputation
39 © LES/PUC-Rio
DRP-MAS (Artificial Intelligence Toolset)
API Bigus*
AI DRPMAS
Forward Chaining
Backward Chaining
Fuzzy Logic
Artificial Intelligence Toolset
Inference Diagnoses
*Bigus, J., Bigus, J., 2001. Constructing Intelligent Agents Using Java, 2nd edition.
40 © LES/PUC-Rio
Performing Diagnosis I/IV
• Goal: to perform diagnosis
• Such analyses are performed based on a set of information provided by the Requester agent (application agent)
Information that can be provided:
• Goal – The goal that was not achieved
• Plan executed – The plan executed by the agent
• Resources: – it may be the case that the resource could not be found, could
not used, the amount was not sufficient, …
• Profile
– The agent’s profile
41 © LES/PUC-Rio
Performing Diagnosis II/IV
Information that can be provided:
• Quality of service
– A degree used to qualify the execution of the plan
• Partners
– The agents with whom the agent has interacted
• Services requested
– Services used by the agents
• Belief Base
– Base of Knowledge
• Devices
– Devices used by the customers.
• Connection
– Type of connection used.
42 © LES/PUC-Rio
Performing Diagnosis III/IV
• The strategy used to make the diagnoses is a hot-spot (flexible point)
• However, the framework provides a set of APIs* to help on the diagnosis:
– backward chaining,
– forward chaining and
– reasoning with fuzzy logic
• The framework provide a default strategy that:
– Compares the amount of resource used and the desired one
– Analyzes the quality of the execution
*Joseph P. Bigus, Jennifer Bigus; Constructing Intelligent Agents Using Java, second edition.
43 © LES/PUC-Rio
Performing Diagnosis IV/IV
• The diagnosis that the default strategy can provide are:
– The wrong amount of resources was used
– Several problems happened at the same time
– It was not possible to identify the problem
44 © LES/PUC-Rio
Providing Recommendations
• The Recommendation agent incorporates the process of advising alternative ways to achieve some goal. It is composed of three steps: (i) to select plans, (ii) to verify the plans need for agents to request information, (iii) to choose good agents
Selecting Plan
Verifying Selected Plans
Choosing agents
45 © LES/PUC-Rio
Scenarios used
• Translation
– Portuguese to English
• Music Market Place
– Buy cd from the name of some music.
Customer
Provider Service
Customer
46 © LES/PUC-Rio
Conclusion
• Two versions of the DRP-MAS
– ASF + Report Framework
– Jadex + Report Framework and Fire model
• Possible Future Works
– Extend the DRP-MAS
• Extend the information set
• Define new strategies of diagnosis and recommendation
• Ubiquitous Computing
– Learning in agents
– Complex scenarios
– Etc.
47 © LES/PUC-Rio
Trabalho Realizados
2004 2006 2007 2008 2009
JADE+BDIJAAF+T
Aplicabilidade do Modelo Belief-Desire-Intention no Java Agent DEveloper Framework
49 © LES/PUC-Rio
Motivação
• Framework JADE não aplica o conceito BDI.
• JADEX não é uma extensão do JADE.
• JADEX e JADE são frameworks totalmente diferentes.
• Agentes JADEX e agentes JADE podem se comunicar a partir de um middleware.
50 © LES/PUC-Rio
Contribuições
• Permitir que se use agentes baseados ou não no modelo BDI. Além de poderem trabalhar em conjunto.
• Representar normas definam quais comportamentos são permitidos e proibidos de serem executados pelos agentes e organizações
• Representar organizações responsáveis por fornecer normas que deverão ser seguidas por agentes de software.
• Permitir definir diferentes políticas de seleção de comportamentos e objetivos para cada agente e organização.
• Manter o framework JADE sendo FIPA-Compliant.
51 © LES/PUC-Rio
Diagrama de Classes
52 © LES/PUC-Rio
Diagrama de Classes (hot-frozen spots)
53 © LES/PUC-Rio
Comparações
• Comparação com Jadex, Jack, Jason, ASF.
• Principais vantagens do JADE+BDI
– FIPA compliant
– Representa BDI
– Aplica mobilidade de agentes
• Vantagens de outros frameworks
– Oferece suporte para organizações e agentes usando o modelo Moise+
– Trabalha com AgentSpeak e Java
54 © LES/PUC-Rio
Considerações Finais
• Estudo de caso Virtual Market Place.
• JADE+BDI ainda não foi utilizado por outros alunos.
• Trabalhos Futuros:
– Melhorar a representação de organizações e normas.
– Aplicar o conceito de papel.
– Disponibilizar framework para uso
– Escrever artigo sobre extensão realizada.
55 © LES/PUC-Rio
Trabalho Realizados
2004 2006 2007 2008 2009
JADE+BDIJAAF+T
JAAF+T: A Framework to Implement Self-Adaptive Agents that Apply Self-Test
57 © LES/PUC-Rio
Motivation
• Self-adaptive systems become one of the main focal points of software engineers.
• Several approaches describing how systems can perform self-adaptation have been investigated.
• It is necessary to test the adapted behavior by investigating its compliance with the new environment requirements at runtime.
58 © LES/PUC-Rio
Motivation
• The main problems of the approaches that test the adapted behavior at runtime are the following:
– it is not possible to define different input data to be used and output assertions to be checked by the tests
– be useful when analyzing the results of the performed tests, different log formats cannot be defined
– the self-test activity that such approaches use is specifically defined to a given self-adaptation process.
59 © LES/PUC-Rio
Goal
• The goal of the presentation is to present an extension of the Java self-Adaptive Agent Framework (JAAF)to apply the self-test.
• We proposed a new control-loop
• A new activity of test is provided
60 © LES/PUC-Rio
AnalyzeDecision
Effector Collect
Test
Self-adaptation Layer
ApplicationApplicationApplicationLayer
New Control Loop with Test Activity
61 © LES/PUC-Rio
Classes related with Collect activity
Class related with Analyze Activity
Class related with Effector Activity
Classes related with Decision Activity
Classes related with Test Activity
62 © LES/PUC-Rio
Test Activity
• The test activity is composed of four steps
– In this step the application designer should relate the actions of the agent to the test cases used to test such actions. In order to define the possible test cases that will be executed, the Test Definition Language (TDL) was created.
63 © LES/PUC-Rio
Test Activity
– The next step defines the data to be used as input data and output assertions while testing the actions. In order to make the definition of such data possible, the Quality Definition Language (QDL) was defined.
64 © LES/PUC-Rio
Test Activity
– After relating the test cases and the actions, and also defining the related data, the tests can be executed when requested by the decision activity. Therefore, the third step of the test activity executes the test per se.
• Different types of test can be executed, such as, unit test, functional test, performance test, etc.
• Nowadays, the framework already provides two types of executions: using JUnit and DBUnit API for unit tests
– In the sequence, it is time to generate the output logs with the resThese logs will be used by the decision activity in order to decide whether to execute the action or choose another one. ults of the executed test.
65 © LES/PUC-Rio
Test Activity
• In order to use the test activity in another control loops, it is only necessary to:
– Define the input and output assertions at design time by using QDL
– Define at design time the types of logs that can be used to format the feedback provided by the test activity by using TDL
– Inform at design time the test cases that will be used to test actions by providing a TDL file
– Implement an activity able to analyze at runtime the logs provided by the test activity. The framework provides as default the Decision class that can be straightforwardly used by the application instance of the framework.
– Call the test activity by providing the action that will be tested.
66 © LES/PUC-Rio
Case Study: Creation of Susceptibility Maps
67 © LES/PUC-Rio
Final Considerations
• Continue studying and proposing approaches that applying self-test concept.
• Writing paper about JAAF-ST
– Self-adaptation + services + self-test
68 © LES/PUC-Rio
Referências
• COSTA, Andrew Diniz da ; SILVA, Viviane Torres da ; LUCENA, Carlos J P . Computing Reputation in the Art Context: Agent Design to Handle Negotiation Challenges. In: Workshop Trust in Agent Societies, in the Seventh International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2008), 2009, Estoril. Proceedigs of the Seventh International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2008), 2009.
• COSTA, Andrew Diniz da ; LUCENA, Carlos J P ; SILVA, Viviane Torres da ; AZEVEDO, S. C. ; AZEVEDO, F. . Art Competition: Agent Designs to Handle Negotiation Challenges. In: Falcone, R.; Barber, S.K.; Sabater-Mir, J.; Singh, M.P. (Org.). (Org.). Trust in Agent Society (TRUST@AAMAS 2008 post-proceedings). Berlim: Springer-Verlag, 2008, v. 5396, p. 244-272.
• S. NETO, B. F. ; COSTA, Andrew Diniz da ; NETTO, M. T. A. ; SILVA, Viviane Torres da ; LUCENA, Carlos J P . A Framework to Implement Self-Adaptive Agents. In: International Conference on Software Engineering and Knowledge Engineering, 2009, Boston. Proceedings of the 21th International Conference on Software Engineering and Knowledge Engineering (SEKE 09), 2009.
69 © LES/PUC-Rio
Referências
• S. NETO, B. F. ; COSTA, Andrew Diniz da ; SILVA, Viviane Torres da ; LUCENA, Carlos J P . JAAF-S: A Framework to Implement Autonomic Agents Able to Deal with Web Services. In: The 4th International Conference on Software and Data Technologies (ICSOFT 2009), 2009, Sofia. Proceedings of the 4th International Conference on Software and Data Technologies (ICSOFT 2009), 2009.
• COSTA, Andrew Diniz da ; SILVA, Viviane Torres da ; ALENCAR, P ; LUCENA, Carlos J P . A Hybrid Diagnostic-Recommendation System for Agent Execution in Multi-Agent Systems. In: ICSOFT-2008, 2008, Porto. Proceedings of the ICSOFT-2008, 2008.
• COSTA, Andrew Diniz da ; Nunes, C.; SILVA, Viviane Torres da; S. NETO, B. F.; LUCENA, Carlos J P . JAAF+T: A Framework to Implement Self-Adaptive Agents that Apply Self-Test. In: The 25th Symposium On Applied Computing (SAC 2010), Switzerland, 2010.
70 © LES/PUC-Rio
Referências
• AZEVEDO, S. C. ; NETTO, M. T. A. ; COSTA, Andrew Diniz da ; Borsato, B.; LUCENA, Carlos J P . Multi-Agent System for Stock Exchange Simulation MASSES. In: IV Workshop on Software Engineering for Agent-oriented Systems (SEAS 2008), 2008, Campinas. Anais do SEAS 2008, 2008.
• COSTA, Andrew Diniz da ; SILVA, Viviane Torres da ; ALENCAR, P ; LUCENA, Carlos J P ; Donald, D. . A Hybrid Diagnostic-Recommendation System for Agent Execution Applied to Ubiquitous Computing Systems. In: IV Workshop on Software Engineering for Agent-oriented Systems (SEAS 2008), 2008, Campinas. Anais do SEAS 2008, 2008.
• COSTA, Andrew Diniz da ; AZEVEDO, F. ; AZEVEDO, S. C. ; SILVA, Viviane Torres da ; LUCENA, Carlos J P . Zé Carioca LES - Agent Reputation Trust (ART) testbed. In: Third Workshop on Software Engineering for Agent-Oriented Systems (III SEAS), 2007, João Pessoa. Anais do III Workshop on Software Engineering for Agent-Oriented Systems (SEAS), 2007.
The End
top related