Transcript 슬라이드 1 - Yonsei

Software Agent
- MAS: multi-agent systems-
Outline
• Definition
• Issues and elements of MAS
–
–
–
–
MAS architectures
Coordination
Collaboration
Several issues in designing competitive MAS
• Applications
• MAS research direction
• Summary
1/28
Multi-agent Systems
• A multi-agent system contains a number of agents…
–
–
–
–
…which interact through communication…
…are able to act in an environment…
…have different “spheres of influence” (which may coincide)…
…will be linked by other (organizational) relationships
• MAS as seen from distributed AI
– A loosely coupled network of entities that work
together to find answers to problems that are
beyond the individual capabilities or knowledge of
each entity
• A more general meaning
– systems composed of autonomous components
that exhibit the following characteristics:
• each agent has incomplete capabilities to solve a
problem
• there is no global system control
• data is decentralized
• computation is asynchronous
2/28
Overview of MAS
• Aspects of multi-agent systems
–
–
–
–
–
–
–
Cooperative vs. competitive
Homogeneous vs. heterogeneous
Macro vs. micro
Interaction protocols and languages
Organizational structure
Mechanism design / market economics
Learning
• Types of MAS
– Cooperative MAS
• Distributed problem solving: Less autonomy
• Distributed planning: Models for cooperation and teamwork
• Typical (cooperative) MAS domains
– Distributed sensor network establishment
– Distributed vehicle monitoring
– Distributed delivery
– Competitive or self-interested MAS
• Distributed rationality: Voting, auctions
• Negotiation: Contract nets
3/28
Comparison with Traditional Approaches
• Traditional
–
–
–
–
Client-server
Low-level messages
Synchronous
Can not do the job!
• Agent
breakthroughs
– Peer-to-peer
topology
– Blackboard
coordination model
– Encapsulated
messaging
– High-level message
protocols
Traditional Software
Client
Function(Parameters)
Server
Return(Parameters)
Agents
Intelligent
Agents
Intelligent
Agents
Intelligent
Agents
Blackboard
Message
Intelligent
Agents
Intelligent
Agents
Intelligent
Agents
Reply
Intelligent
Agents
Intelligent
Agents
Intelligent
Agents
A14<MAS>-4
4/28
Main Points in MAS
• MAS researchers develop communications languages, interaction
protocols, and agent architectures that facilitate the development of
multi-agent systems
• MAS researcher can tell you how to program each ant in a colony in
order to get them all to bring food to the nest in the most efficient
manner, or how to set up rules so that a group of selfish agents will
work together to accomplish a given task
• MAS researchers draw on ideas from many disciplines outside of AI,
including biology, sociology, economics, organization and
management science, complex systems, and philosophy
5/28
Key Elements of MAS
• A coordination mechanism supported by a common agent
communication language and protocol
• A collaboration mechanism supported by agent community
architecture (including agent and interaction architecture) to support
the organization goal
• A shared ontology
• Popular MAS architectures
–
–
–
–
–
Object Manager Group (OMG)
Foundation for Intelligent Physical Agents (FIPA)
Knowledgeable Agent-oriented System (KAoS)
Open Agent Architecture (OAA)
General Magic group
6/28
MAS Architectures (1)
• OMG’s Model
– Composed of agents and agencies that collaborate using general patterns and
policies
– Agents are characterized by: capabilities, type of interaction and mobility
– Agencies support:
• concurrent execution of agents
• security
• agent mobility
• FIPA’s Model
–
–
–
–
–
–
Agents
Agent Platform (AP)
Directory Facilitator (DF)
Agent Management System (AMS)
Agent Communication Channel (ACC)
Agent Communication Language (ACL)
7/28
MAS Architectures (2)
• KAoS’s Model
– An Open Distributed Architecture for Software agents
– Defines various agent implementations
– Uses conversation policies to elaborate on agent-to-agent communication
• OAA Model
8/28
MAS Architectures (3)
• General Magic’s Model
– A commercial agent technology for electronic commerce
– Views MAS as an electronic marketplace
– The marketplace is modeled as a network of computers supporting a collection of
places that offer services to mobile agents
– The mobile agents:
• can travel, meet other agents, create connections to other places
• they have authority
Agent Name Server
Address
Book
• Zeus: a MAS development toolkit
request
Co mmon Message Format (Language)
Shared mesage content
representation and ontology
reply
Agent
Facilitator
B
A
D
MESSAGE
Transport Protocol
C
Agent
Agent
Agent
Perform
Task A
Perform
Task C
Perform
Task D
Abilit ies
Database
External
program
9/28
MAS Architectures (4)
• Geo-Agents (GIS agents) Architecture
Other Agent Systems
User
Task(GeoScrip
t)
UI AgentReply
Query agent
Geo-Agents
Administrator
Exchange registry
Query agent
Query agent
Facilitator
Query agent
Pass task
Reply
Coordinate Coordinate
Task Agent
Domain (Service) Agent
Control/Reply
Task Agent
Domain (Service) Agent
Collaborate
Retrieve
Collaborate
Data sources
10/28
Coordination
• Coordination: a process to manage dependencies among activities
• Three aspects of coordination
– Activity aspect
• What activity to execute?
• When an activity should be executed?
• Model to coordinate distributed tasks: Statecharts, Flowcharts, Process algebra, Lotos,
SDL, Estelle …
– Conversation (state) aspect
• What is the structure of the conversation among the coordinating entities?
• FSM, Petri-Nets, State Transition Diagrams
– Implementation aspect
• How to implement distributed software systems where software components coordinate
their actions
11/28
KQML
Coordination
• Knowledge Query and Manipulation Language (KQML) is both a
message format and a message-handling protocol to support run-time
knowledge sharing among agents
(ask-all
/* message layer */
• KQML comprise a substrate :content
on which"price(IBM,
to develop [?price,
higher-level
models
?time])“
of inter-agent interaction such as contract
nets layer */
/* content
:receiver stock-server
• KQML is a coordination mechanism from
the conversation
/* communication
layeraspect
*/
:language standard_prolog
NYSE-TICKSwhich defines the
• KQML contains an extensible:ontology
set of performatives,
:sender
me)may use
permissible speech acts
agents
• Example performative:
12/28
KQML: Types of Performatives
•
•
•
•
•
•
•
•
•
•
•
Coordination
Basic informative performatives: tell, deny, …
Database performatives: insert, delete, …
Basic responses: error, sorry, …
Basic query performatives: ask-one, ask-all, evaluate,…
Multi-response query performatives: stream-all, …
Basic effector performatives: achieve, …
Generator performatives: standby, ready, next, …
Capability-definition performatives: advertise
Notification performatives: subscribe
Networking performatives: register, forward, pipe, broadcast, …
Facilitation performatives: broker-one (all), recommend-one (all),
recruit-one (all)
13/28
Collaboration
• Collaboration refers to cooperative effort among agents to reach a
single goal by exchanging knowledge built upon the underlying
coordination mechanism
• Example mechanism: Contract Net Protocol (CNP)
– Negotiation as a collaboration mechanism
– Negotiation on how tasks should be shared
• A task (plan) may be decomposed in a hierarchy of subtasks (hierarchical planning)
• An agent may subcontract another agent to perform a (sub)task.
agent
agent
Bid
Contract
14/28
Collaboration
Phase 1: Task Announcement
Task announcement
("broadcast")
- The contractor agent publicly
announces a task.
- Potential candidates evaluate
the task according to their wo
n skills and availability.
Contractor
Potential candidate agents
Phase 2: Submission of Bids /
Proposals
Bid
Bid
- Agents that satisfy the requir
Contractor
Candidate
Candidate
emenst, i.e., are able to perfor
m the task, send their bid / pro
posal to the contractor.
15/28
Collaboration
Phase 3: Selection
- The selection of the best can
didate is made by the contract
or based on received bids and
on the CVs of the candidates.
Contractor
Selected
candidate
Phase 4: Contract awarding
Contract
- A contract is established betwe
en the contractor and the selecte
d candidate.
Contractor
Contracted
agent
- A privileged bilateral communic
ation channel is established betw
een the two agents.
16/28
Several Issues in Designing Competitive MAS
• Distributed rationality
• Pareto optimality
• Stability
17/28
Distributed Rationality
Competitive MAS
• Techniques to encourage/coax/force self-interested agents to play
fairly in the sandbox
–
–
–
–
Voting: Everybody’s opinion counts (but how much?)
Auctions: Everybody gets a chance to earn value (but how to do it fairly?)
Contract nets: Work goes to the highest bidder
Issues:
•
•
•
•
Global utility
Fairness
Stability
Cheating and lying
18/28
Competitive MAS
Pareto Optimality
• S is a Pareto-optimal solution iff
–  S’ ( x Ux(S’) > Ux(S) →  y Uy(S’) < Uy(S))
– i.e., if X is better off in S’, then some Y must be worse off
• Social welfare, or global utility, is the sum of all agents’ utility
– If S maximizes social welfare, it is also Pareto-optimal (but not vice versa)
Which solutions
are Pareto-optimal?
Y’s utility
Which solutions
maximize global utility
(social welfare)?
X’s utility
19/28
Competitive MAS
Stability
• If an agent can always maximize its utility with a particular strategy
(regardless of other agents’ behavior) then that strategy is dominant
• A set of agent strategies is in Nash equilibrium if each agent’s
strategy Si is locally optimal, given the other agents’ strategies
– No agent has an incentive to change strategies
– Hence this set of strategies is locally stable
• Prisoner’s dilemma
– Pareto-optimal and social welfare maximizing solution: Both agents cooperate
– Dominant strategy and Nash equilibrium: Both agents defect
Cooperate
Defect
Cooperate
3, 3
0, 5
Defect
5, 0
1, 1
20/28
Development of MAS
• Define the organization of the MAS according to the problem
specification (or solution structure)
• Decide the coordination mechanism
• Select a MAS implementation framework, e.g., Zeus, that supports
the coordination mechanism
• Implement the collaborative mechanism which support the MAS
organization
• Implement shared ontology
• Implement each task agent (including customizing associated
communication module)
• Customize middle agents
–
–
–
–
–
Facilitators
Mediators
Brokers
Matchmakers and yellow pages
Blackboards
21/28
Applications of MAS
• Advanced Manufacturing Management Systems
– Agents as representatives of machines, users, business processes, etc.
• Intelligent Information Search on Internet
– Some agents may show learning capabilities (learn the preferences of their
users, ..)
• Intelligent security enforcement on Internet
– Agents are representative of sensors or IDSs
• Shopping Agents in Electronic Commerce
– With search, price comparison, and bargaining capabilities
• Multi-agent auction in E-commerce
• Distributed Surveillance
– For information search or to look for special events informing their users of relevant
news
• Distributed Signal Processing
– For problem diagnosis, situation assessment, etc. in the network
• Distributed Problem Solving
– Collaborative design, scheduling, and planning
22/28
Agent Organizations
MAS Research
Directions
• Multiple (human and/or artificial) agents
• Goal-directed (goals may be dynamic and/or conflicting)
• Affects and is affected by the environment
• Has knowledge, culture, memories, history, and capabilities (distinct
from individual agents)
• Legal standing is distinct from single agent
• Q: How are MAS organizations different from human organizations?
23/28
Organizational Structures
MAS Research
Directions
• Exploit structure of task decomposition
– Establish “channels of communication” among agents working on related subtasks
• Organizational structure:
– Defines (or describes) roles, responsibilities, and preferences
– Use to identify control and communication patterns:
• Who does what for whom: Where to send which task announcements/allocations
• Who needs to know what: Where to send which partial or complete results
24/28
Communication
MAS Research
Directions
• Communication models
– Theoretical models: Speech act theory
– Practical models:
• Shared languages like KIF, KQML, DAML
• Service models like DAML-S
• Social convention protocols
• Communication strategies
– Connectivity (network topology) strongly influences the effectiveness of an
organization
– Changes in connectivity over time can impact team performance:
• Move out of communication range  coordination failures
• Changes in network structure  reduced (or increased) bandwidth, increased (or reduced)
latency
25/28
Learning in MAS
MAS Research
Directions
• Emerging field to investigate how teams of agents can learn
individually and as groups
• Distributed reinforcement learning
– Behave as an individual, receive team feedback, and learn to individually contribute
to team performance
– Iteratively allocate “credit” for group performance to individual decisions
• Genetic algorithms: Evolve a society of agents (survival of the fittest)
• Strategy learning: In market environments, learn other agents’
strategies
26/28
Adaptive Organizational Dynamics
MAS Research
Directions
• Potential for change:
– Change parameters of organization over time
– That is, change the structures, add/delete/move agents, …
• Adaptation techniques:
–
–
–
–
–
Genetic algorithms
Neural networks
Heuristic search / simulated annealing
Design of new processes and procedures
Adaptation of individual agents
27/28
Summary
• “Agent” means many different things
• Different types of “multi-agent systems”:
– Cooperative vs. competitive
– Heterogeneous vs. homogeneous
– Micro vs. macro
• Lots of interesting/open research directions:
–
–
–
–
Effective cooperation strategies
“Fair” coordination strategies and protocols
Learning in MAS
Resource-limited MAS (communication, …)
• Next lecture
– Communication & Platform
28/28