intelligent agent characterization and uncertainty management with fuzzy set theory a tool to...
TRANSCRIPT
Intelligent Agent Characterization and Uncertainty Management with
Fuzzy Set Theory
A Tool to Support Early Supplier Integration Journal of Intelligent Manufacturing (1999) 10
Author : Pamela McCauley-Bell
Intelligent Agent
Outline
IntroductionArchitectural InnovationArchitectural Innovation through Early
Supplier Integration Intelligent AgentsFuzzy Set Theory ImplementationConclusion
1. Introduction
Intelligent manufacturing systemSome important computational toolsAgent theoryUncertainty management
Intelligent Manufacturing System
The field of intelligent manufacturing system :
Simulation
Virtual reality
Robotics
Some important computational tools :
Personal computer (PC)
Knowledge-based systems (KBS)
Neural networks
Programming logic controllers (PLC)
Intelligent agents (IA)
Computational Tools
Agent theory is concerned with the question of what an agent is , and the use of mathematical formalisms for representing and reasoning about the properties of agents.
Agent Theory
Uncertainty Management
Uncertainty representation Uncertainty managementUncertainty measurementUncertainty modeling
Uncertainty Management
Uncertainty management approaches :
Probability model
Dempster-Shafer theory
Certainty factors
Fuzzy set theory
4. Intelligent Agents
Agent definitionAgent categoriesAgent classificationsAgent CharacteristicsHeterogeneous agent systems
Agent definition
Agents are referred to the assumption of autonomy.
Autonomy is a basic consideration because the entity is expected to function with a mission and without the benefit of continuous supervision or guidance.
Agent definition
An Autonomous agent can be defined as a system situation within, and existing as a part of an environment. The agent senses the predefined parameters and actors on them, over time, in pursuit of its own agenda to effect what it senses in the future.
Agent categories
Biological agents
Robotic agents
Computational agentsArtificial life agents
Software agents
Task specific agents
Entertainment agents
Viruses agents
Agents classifications
Collaborative agentsInterface agents
Mobile agents
Reactive agents
Information/internet agents
Hybrid agentsSmart agents
Agent CharacteristicsAutonomyCommunication ability (Mobility)Capacity for cooperation
Rule-based
Capacity for reasoning (Intelligence)
Knowledge-basedArtificial evolution-based
Adaptive behaviorTrustworthiness
Heterogeneous agent systems
To create the service for a wide range domains, a key requirement for interoperation among heterogeneous agents is having an agent communication language that permits agents of varying origins to communicate with each other.
Heterogeneous agent systems
Three important questions: What is an appropriate agent communication language?
How are agents capable of communicating in this language constructed?
What communication architectures are conductive to cooperation?
Fuzzy Set Theory
Fuzzy modeling approach is more consistent with human information processing, because humans have an ability to analyze imprecise concepts, which are not thoroughly understood or quantified.
The “imprecise” knowledge is essential to human-cognitive processes and is effectively modeled through the use of linguistic values and degrees of membership in fuzzy set theory(FST).
FST deals with the imprecision associated with many variables by permitting a grade of membership to be defined over the interval [0,1].
Mathematical Definition of Fuzzy Sets
Consider a finite set of objects X, define the finite set as :
X = x1 , x2,…, xn
where xi are elements in the set X.
Each element, xi, has a particular membership value, ui, which represents its grade of membership in a fuzzy set.
A fuzzy set A can be represented as a linear combination of the following form:
A = u1(x), u2(x) ,…, un(x),
The methods of dealing with uncertainty in decision support
Probability : Bayes’ theorem
Probability intervals : Dempster-Shafer theory
Fuzzy variables
Probability : Bayes’ Theorem
In non-numerical combination algorithms that merge multiple hypothesis data vectors from N sensors, the sensors that are used to make the decisions include statistical classifiers that classify measured features into a vector of parameters.
For the Bayesian inference case, these parameters are forward-conditional probabilities. Bayes’ rules is applied to compute a composite, a posteriori probability.
The maximum a posteriori (MAP) is applied to select the most likely hypothesis.
Bayes’ Theorem
The law of total probability
P(C) = P(C1)P(C | C1) + P(C2)P(C | C2) +…+ P(Ck)P(C | Ck)
=
Bayes’ theorem
P(Cj | C)=
k
i
ii CCPCP1
)|()(
k
i
ii
jjj
CCPCP
CCPCP
CP
CCP
1
)|()(
)|()(
)(
)(
P(Cj ) is called prior probability of Cj ,
P(Cj | C) is called posterior probability.
Probability Intervals : Dempster-Shafer Theory
The Dempster-Shafer theory of uncertainty attempts to distinguish between ignorance and certainty.
This model permits
P(A) + P(B) 1 ≦
where the P(A) and P(B) represents the strength of evidence or confidence.
Probability Intervals : Dempster-Shafer Theory
Based on identifying the believability of a function or proposition, the function f represents the measure of belief committed to a given proposition or piece of sensory information.
Each hypothesis is represented by two parameters: supportability and a plausibility variable.
( 0 S(X) 1 )which describe the degree to which measurement support the hypothesis;
(2) a plausibility variable
( 0 P(X) 1 )which represents the degree to which the evidence fails to refute the hypothesis.
(1) supportability
Probability Intervals : Dempster-Shafer Theory
The difference in plausibility and supportability
D(I) = P(x) - S(x)
is a measure of ignorance about the hypothesis.
Dempster’s rule of combination, analogous to Bayes’ rule, provides a means of computing composite supportability / plausibility intervals (credibility intervals) for each hypothesis, reducing the uncertainty in the measured data.
Probability Intervals : Dempster-Shafer Theory
Fuzzy Variables
Fuzzy set theory manages uncertainty in decision making by representing the level of certainty in a proposition through a membership function.
Three important principles for managing uncertainty are:
minimal uncertainty maximal uncertainty minimal invariance
Implementation
The process for the development of fuzzy intelligent agents:
A. Evaluate the needs of the environment and required agents.
B. Identify the role of agents.
C. Describe the personification of agents.
D. Characterize the types of uncertainty.
E. Define agents’ interaction.
Identifying the Roles of Agents
Missionaries : work to deliver information and attempt to reach or convert as many agents as possible by giving them their information.
Brokers : represent objects and negotiate for objects, can change belief function of an object.
Dispatcher :distributes information, gives instructions; capable but not required to store information for learning.
Scout :sound alarm if something appears wrong.
Controller: brains of community mayor; capable of storing information and learning.
Identifying the Personality
Personality : used to simulate a community of individuals interacting.
Aggressiveness
Vulnerability
Attractiveness
Truth constant
Aggressiveness
It is used to define the fervor with which a given agent is expected to pursue to other agents.
The degree of aggressiveness of an agent is effected by the number of agents (x) that the given agent direct access or links to in the model.
If x is larger, then the agent is considered aggressive because it has the ability to influence a large variety of agents in the system.
The membership function for aggressiveness is a s-shape curve.
Vulnerability
A function of the basic personality and the number of incoming source (y) or agents that have the potential to influence or reach the agent.
If the agent only contacts with one of these agents, the information which the agent possesses is untrustworthy.
The membership function for vulnerability is an increasing s-shape curve.
It is also a function of the basic personality, but may be enhanced or reduced by the agent interaction or input-to-output ratio (z).
The smaller the ratio is, the more attractive the agent is considered.
The membership function for attractiveness is an decreasing s-shape curve.
Attractiveness
It is used to evaluate the nature of the environment. The truth constant is an integer value on the interval [1,5].
A low-truth constant is associated with a hostile environment indicating that the integrity of the information can not be certified due to parameters beyond the systems control or knowledge.
The overall belief function will be a function of the aggressiveness, attractiveness, vulnerability, and truth constant.
Truth Constant
The categories and quantification of uncertainty
The types of uncertainty:
Non-specificity (imprecision)
Fuzziness (vagueness)
Strife/discord
Nonspecificity
Nonspecificity : Vi = [U1, U2 ,…, Un], and this type of uncertainty is manifested when two or more alternatives are left unspecified. This may be a result of variety, generality, diversity, equivocation.
Nonspecificity
Hartley function (U) provides an effective method to quantify the uncertainty. For any non-empty fuzzy set A defined on a finite universal set X, the generalized Hartley function has the form:
||log)(
1 )(
0 2A
Ah
Ah
U(A) =
where denotes the cardinality of the -cut of A and h(A) is the height of A.Observe that U(A), which measures nonspecificity of A, is a weighted average of values of the Hartley function for all distinct -cut of the normalized counterpart of A, defined by A(x)/h(A) for all x X.
|| A
Discord/strife
Discord /strife : Di = [D1, D2 ,…, Dn], is the type of uncertainty characterized by disagreement in choosing among alternatives and may result from dissonance, incongruity, discrepancy, and conflict.
The value will be obtained by taking the general form of the union, which takes the largest membership value contained within the set to represent the union.
Fuzziness
Fuzziness : Xi = [F1, F2 ,…, Fn], is characterized by lack of definite or sharp distinction among alternatives and may result from vagueness or any variety of indecisiveness.
Fuzziness will be obtained by using the Hamming distance.
In general, a measure of fuzziness is a function
where denotes the set of all fuzzy subsets of X (fuzzy power set)
For each fuzzy set A, this function assigns a non-negative real number of f(A) that expresses the degree to which the boundary of A is not sharp.
RXf )(:)( X
Relationship between personality traits and measures of uncertainty
The personality traits and the measures of uncertainty are metrics used to describe the capabilities and truthfulness associated with an agent and its information.
The personality traits provides an indication of the behavior and the interactability of the agent with other community members.
The measures of uncertainty produces a measure of reliability for the task that the agent has performed.
Agent Interaction
The initiation of activity in the fuzzy-agent environment begins with a trigger event.
Trigger events may be defined as the presentation of information and may be produced from the activity of a related event, acquisition of information, given set of instructions, or the lapse of a pre-defined time period.
Agent Interaction
The trigger event for each agent is different and is code into the rule base.
The process is continuous, but a task within the process ends with the termination or delivery of a specific piece of information.
In the initial model, all agents are not free to interact with each other.
Example
The hypothetical example consist of a prime and three first tier suppliers.
The developing steps:
1. Identify prime and the first tier suppliers.
2.identify significant second tier suppliers.
3.Define the bank of information, or the necessary inputs.
Example
4.Identify types of agent roles.
. Provide personality,
. Provide truth constant,
5.Provide guidelines for agent interaction
6.Define trigger events
7.Design electronic blackboard to display the information.
Fig. 1. Structure for early supplier integration network.
Example
Prime
IA Team 1 IA Team 2 IA Team 3
Supplier 1 Supplier 2 Supplier 3
SS11 SS12 SS21 SS22 SS31 SS32
Table 2. Team of agents to support early supplier integration
Primary personality
Primary personality
Primary personality
Primary personality
Missionary
Scout
Runners
Time keepers
Aggressive
Aggressive
Aggressive
VulnerableAttractive
To deliver information to the prime and all relevant suppliers
Search for violation of constraints
Deliver information to the primes database
To deliver information to the prime and all relevant suppliers
Continuous: Ongoing dissemination of information to predefined sources
Violation of a system constraint
Continuous
A modification that produces a change in the active product schedule
Example
Conclusion
Produces a framework for the development of fuzzy generic agents;
Uses the fuzzy agents to manage uncertainty in product development system;
Provides continuous interaction among suppliers and the prime;
The model can be easily translated to model the needs of intelligent industries process.