300 likes | 313 Views
Explore the role of agent technology in e-learning systems with a focus on goal modeling and goal-oriented action selection. Learn about Agent Mediate Service-Oriented Grid, Goal Net modeling, and characteristics of Agent's Goals.
E N D
Agent Mediated Grid Services in e-Learning Chun Yan, Miao School of Computer Engineering Nanyang Technological University (NTU) Singapore 639798 April, 2004
Preface: Agent • Agent is an emerging new paradigm for next generation e-Systems in various domains. • Agent technology is identified by “MIT Technology Review” as one of the technologies that will change the world”. • It is predicted that in 10 years time, most new software developments will contain embedded agent systems.
2020 Vision: How People Learn • The familiar “world to the desktop” • “Alice in Wonderland”: computer based agents assist learners in diverse ways • “ubiquitous learning”: embedded agents in handheld wireless devices and in real objects i.e. intelligent objects
Preface: Goal Orientation • An agent is defined as the one who acts on behalf of human beings. • By nature, a human being does things based on goals. Goal orientation is a key character of agents. • A large majority of the current efforts on agent modeling and development still employ object-oriented methodologies, which model an agent as an extended object • Agents are goal oriented, which necessitates a shift in modelling paradigm, from object-oriented modeling to goal-oriented modeling.
Agenda • Agent Mediate Service Oriented Grid • Goal Net: A Goal oriented modeling approach to agent oriented systems • Goal Selection and Action Selection • Modeling of MAS with Goal Net • Goal Autonomous Agent • Agent Mediate Grid Service in e-Learning
Agent’s Goal Model • Task Oriented: an agent lives in a task-oriented domain; the goal of an agent is a set of tasks to perform. • State Oriented: an agent lives in the state-oriented domain. The agent’s environment is evolved with a finite set of states. A goal of an agent is a desired state that the agent tries to reach from its current state by going through a sequence of states.
Characterizing Agent’s Goal • To model the complex goals of agents, a characterization of an agent’s goal with different properties such as composite goal, fuzzy goal, partial goal, sub goal etc. is highly needed. • To enable agents to present such goal characters, new goal models are demanded.
Goal Net: Overview • A Goal Net is composed of five basic objects: states, transitions, arcs, branches, and tokens. composite state branch token transition transition arc atomic state composite state
Goal Net: State • State • State is a system situation at a time during agent running • Atomic, Composite • Goal is a desired state that an agent intents to reach. • In a Goal Net, a composite state is a goal.
Direct Transition Conditional Transition Probabilistic Transition Fuzzy Cognitive Transition Goal Net: Transition • Transition • Defines actions to transit from one state to another state. • Defines action selection mechanism • Direct, Conditional, Probabilistic, Fuzzy
Sj Si Sj Sk Si Sequence Concurrency Sj Sj Si Sk Sk Si Choice Synchronization Goal Net: Transition • Transitions can represent four basic relationships between states: sequence, conflict, concurrency, and synchronization.
Goal Net: Arc,Token,Branch • An arc is used to connect a state to a transition or a transition to a state. It indicates the relationship between the state and the transition it connects. • A token is used to indicate agent’s current activities in different states. It presents dynamic behaviors of the goal model. It indicates the progress of the goal pursuit process. • The branches are used to represent the decomposition of a composite state.
Goal Net: Measurement • Goal Measurement • Achievement: represents a recognizable benefit of reaching a goal; • Distance: indicates how close the current state is to a composite state or a sub goal; • Completeness: represents a percentage of the entire goal fulfillment; • Cost: means the time, memory, money, etc. spent or required to be spent from one state to another.
Goal Net: Goal/Action Selection • Goal Selection – Goal Autonomy • Take future goals/stats into consideration • Achievement, Cost, Constraint, Trust, Index • Action Selection – Behavior Autonomy • Sequential execution • Situation Action: Rule-based inference • Probabilistic inference • Fuzzy Cognitive Inference
Goal Net: Action Selection • Sequential execution: This is the simplest situation. There is no action selection needed. Agents can move from one state to the next state by the execution of the fixed sequence of actions. • Rule-based inference: In this situation, complete information for action selection is present. Agents can make decision according to the rules and current values of all the factors or states.
Goal Net: Action Selection • Probabilistic inference: In this situation, information for action selection is not complete. A Bayesian network that represents the relationships between factors and actions can be constructed. An agent then reasons its actions through the Bayesian network inference. • Fuzzy Cognitive Inference: A Fuzzy Cognitive map that represents the relationships between factors and actions can be constructed. An agent reasons its actions through fuzzy cognitive inference.
Modeling MAS with Goal Net • In addition to an agent goal model, Goal Net also serves as a goal-oriented requirement and modeling tool, and a multi-agent identification, organization and coordination model. • From Goal Hierarchy to Agent Hierarchy • Agent Identification • Agent Coordination • Goal Net is able to assist in whole life cycle for development of agent-oriented applications
Agent Mediated Service Oriented Grid 7. End Users 6.Consumer Applications Applications Applications A A 5. Service Agents A A A A 4. Information Service Center A 3. Marketing Service Agents A A A A 2. Grid Services Services Services 1. Provider Applications Applications Applications
Provide Service Received Request Process Request Sent Results Exception Processed Negotiated Job Dispatched Marketing Agents in Agent Grid
Obtain Service Request Received Service Located Serviced Services Discovered Service Selected Negotiated Request Sent Search Service Query Prepared Services Found Broadcast Service Agents in Agent Grid
From Goal Hierarchy to Agent Hierarchy: MAS Derivation Marketing agent X Process request agent Y Negotiate agent Z A Service agent B Locate service agent Lookup agent C D Negotiation agent
Goal Autonomous Agent • The agent whose goal is modeled with the Goal Net is able to present both behavior autonomy and goal autonomy in a dynamic changing environment. We call this type of agents goal autonomous agents.
Goal Autonomous Agent Life Cycle • PR2A • Perceive: The agent perceives its environment continuously to sense any new situations. • Reason for goal selection: The agent infers the next goal, based on its goal model, knowledge, and the perception of its environment. • Reason for action selection: The agent infers actions based on the selected goal, knowledge, and the perception of its environment. • Act: The selected actions are executed.
Database Knowledge Base Data Goal Knowledge Inference Engine perception action Environment Environment controller communication Goal Autonomous Agent • Agent Model
E-Learning System E-learning service Learner preparation Learning Pre-Assessment Teaching Learning Object Delivery Post-Assessment Learning Path Generated Role Selected
Learning service agent Courseware servers in a Grid environment Learning agent Learner preparation agent E-Learning Grid Services System
Learning service agent Courseware servers in a Grid environment Learning agent Learner preparation agent E-Learning Grid Services System
Conclusion • Goal Net serves as a goal-oriented modeling and analysis tool, an agent goal model, and a multi-agent modeling, identification and organization model. • As a new agent goal model, Goal Net enables the agents to present both behavior autonomy and goal autonomy. • The modeling and design of goal autonomous multi-agent systems using Goal Net have demonstrated a promising approach for designing and developing intelligent, open distributed agent systems in grid service in e-Learning.