* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download Components of KBS
Survey
Document related concepts
Soar (cognitive architecture) wikipedia , lookup
Technological singularity wikipedia , lookup
Ecological interface design wikipedia , lookup
Incomplete Nature wikipedia , lookup
Expert system wikipedia , lookup
Personal knowledge base wikipedia , lookup
Intelligence explosion wikipedia , lookup
Embodied cognitive science wikipedia , lookup
Ethics of artificial intelligence wikipedia , lookup
Philosophy of artificial intelligence wikipedia , lookup
Existential risk from artificial general intelligence wikipedia , lookup
Transcript
Architecture of KBS Components of KBS Components of KBS Notes We have to build systems with specific characteristics: Problem solving using symbolic information Resolution by mean of reasoning and heuristic methods Capacity to explain the resolution process Interactivity (with a user/the environment) Able to adapt to the environment A basic set of components is needed Reasoning subsystem Knowledge storage subsystem (Knowledge base) Knowledge use and interpretation subsystem (solving mechanism) Problem state storage subsystem (working memory) Explanation and resolution inspection subsystem Communication and interface subsystem Learning subsystem C \ BY: $ (LSI-FIB-UPC) Artificial Intelligence Architecture of KBS Term 2009/2010 1 / 24 Components of KBS Components of KBS Notes Reasoning subsystem Knowledge storage Subsystem Learning subsystem Explanation and inspection subsystem Problem state Knovledge use and interpretation subsystem storage subsystem Comunication and interface subsystem C \ BY: $ (LSI-FIB-UPC) Artificial Intelligence Architecture of KBS Term 2009/2010 2 / 24 KBS based on production systems KBS based on production systems Notes Problems are solved using the reasoning mechanism of an inference engine The domain knowledge is described by means of an ontology The problem solving knowledge is described using production rules or an equivalent formalism C \ BY: $ (LSI-FIB-UPC) Artificial Intelligence Term 2009/2010 3 / 24 Architecture of KBS KBS based on production systems Knowledge storage subsystem Notes It will store all the knowledge needed to solve problems in the domain of application We can find three kinds of knowledge: Factual knowledge (domain objects and their characteristics) Relational knowledge (relations among the domain objects) Conditional knowledge (Deductive knowledge about the problem) The two first kind of knowledge are described in the domain ontology The third one represents the knowledge related to the resolution of the problem and is described by production rules C \ BY: $ (LSI-FIB-UPC) Artificial Intelligence Architecture of KBS Term 2009/2010 4 / 24 KBS based on production systems Knowledge storage subsystem: Rules Notes Conditional knowledge includes: Deductive knowledge (structural): Describes the process of problem solving as a chain of deductions Knowledge about goals (strategic): Drives the resolution process to the solution Causal knowledge (supportive): Supports explanation of the problem resolution process Rule modules: Sets of related rules Eases the development and maintaining of the system Increases the efficiency of the reasoning process Allows to implement strategies for the use of the knowledge (meta-knowledge, meta-rules) C \ BY: $ (LSI-FIB-UPC) Artificial Intelligence Architecture of KBS Term 2009/2010 5 / 24 KBS based on production systems Knowledge storage subsystem: Meta-Rules Notes They describe high level knowledge about the process of solving the problem They allow to take control of the resolution process Activating and deactivating rules/modules Deciding the application order of rules/modules Deciding resolution strategies, handling of exceptions, uncertainty, ... This knowledge is more difficult to obtain from the experts C \ BY: $ (LSI-FIB-UPC) Artificial Intelligence Term 2009/2010 6 / 24 Architecture of KBS KBS based on production systems Knowledge use and interpretation subsystem Notes Usually is an inference engine It applies its execution cycle to solve the problem Detection of applicable rules Selection of the best rule (Domain independent strategy or driven by metaknowledge) Execution of the rule C \ BY: $ (LSI-FIB-UPC) Artificial Intelligence Architecture of KBS Term 2009/2010 7 / 24 KBS based on production systems Problem state storage subsystem Notes Stores the problem initial facts and all the facts obtained during the resolution process It can store also other kinds of information needed for the control of the resolution process or other subsystems Order of deduction of the facts Use preferences of the facts Rule that produced each fact Rules recently fired Backtracking points ... C \ BY: $ (LSI-FIB-UPC) Artificial Intelligence Architecture of KBS Term 2009/2010 8 / 24 KBS based on production systems Explanation and resolution inspection subsystem Notes The possibility to justify the decisions taken during the resolution process gives credibility to the system It allows also to detect wrong assumptions or decisions during the process A system should be able to answer why and how Different levels of explanation: Trace: A list of the steps of the resolution is given Justification: The reason why each element that appear in the trace of the resolution is given (reasoning path, questions, facts, preferences, subgoals, ...) C \ BY: $ (LSI-FIB-UPC) Artificial Intelligence Term 2009/2010 9 / 24 Architecture of KBS KBS based on production systems Learning subsystem Notes Usually the set of problems that a KBS solves is closed In some domains it is necessary to adapt to the environment and to solve new problems Learning can happen: During the development process of the KBS: The knowledge engineering process is substituted or complemented by inductive learning methods, building a model of the problem from examples During the resolution process: Incorrect solutions are detected and corrected, new rules that make more efficient the resolution process are learned C \ BY: $ (LSI-FIB-UPC) Artificial Intelligence Architecture of KBS Term 2009/2010 10 / 24 KBS based on Case Based Reasoning Cased Based Reasoning (CBR) Notes The solution of a problem is obtained identifying a previous similar solution Advantages: The problem of knowledge acquisition is reduced It is easier to maintain/correct/extend the system The solving process is more efficient It allows to obtain explanations closer to the user experience C \ BY: $ (LSI-FIB-UPC) Artificial Intelligence Architecture of KBS Term 2009/2010 11 / 24 KBS based on Case Based Reasoning Execution cycle Notes It has four phases 1 Retrieval: The more similar stored cases are retrieved 2 Reuse: The solution of the cases are obtained 3 Revision: The retrieved solution is evaluated and adapted 4 Retention: The relevance of the new case is assessed and the case is stored if it is an interesting one C \ BY: $ (LSI-FIB-UPC) Artificial Intelligence Term 2009/2010 12 / 24 Architecture of KBS KBS based on Case Based Reasoning Execution cycle Notes Re tri e va l New Case Case Stored Re ten tio n Case Retrie ved New Case Cases Domain Knowledge R eu se Case Revised Revis C \ BY: $ (LSI-FIB-UPC) ion Case Solved Artificial Intelligence Architecture of KBS Term 2009/2010 13 / 24 KBS based on Case Based Reasoning Knowledge storage subsystem Notes The knowledge is composed by cases A case is a complex structure (characteristics, solution) The cases are stored in the case base (structure, indexing) We will have also knowledge about: How to evaluate case similarity How to combine/adapt/retrieve solutions How to evaluate solutions C \ BY: $ (LSI-FIB-UPC) Artificial Intelligence Architecture of KBS Term 2009/2010 14 / 24 KBS based on Case Based Reasoning Knowledge use and interpretation subsystem Notes It is based on the execution cycle of Cased Based Reasoning Retrieve the more similar cases from the case base Obtain the solution from the cases Combine/adapt the solution (procedures/reasoning) C \ BY: $ (LSI-FIB-UPC) Artificial Intelligence Term 2009/2010 15 / 24 Architecture of KBS KBS based on Case Based Reasoning Problem state storage subsystem Notes Information of the current case Most similar cases retrieved Reasoning results from evaluating/combining/adapting the solutions C \ BY: $ (LSI-FIB-UPC) Artificial Intelligence Architecture of KBS Term 2009/2010 16 / 24 KBS based on Case Based Reasoning Explanation - Learning Notes Explanation It is a part of the information of a case This explanation will be complemented by the reasoning process to combine/adapt the solutions Learning To add new cases (easier than in production systems) The new case has to be different enough (evaluation) Cases could be forgotten (low usage, similar to others) C \ BY: $ (LSI-FIB-UPC) Artificial Intelligence Architecture of KBS Term 2009/2010 17 / 24 Other methodologies Other methodologies Notes Systems based on neural networks Intelligent Agents/Multiagent systems C \ BY: $ (LSI-FIB-UPC) Artificial Intelligence Term 2009/2010 18 / 24 Architecture of KBS Other methodologies Neural networks Notes Conexionist artificial intelligence The base element is the neuron (computational element) Neuron: Inputs, outputs, state, functions for the combination of inputs and state and function to compute the output Neurons are organized in networks with layers A neural network associates inputs (problem description) to outputs (solution of the problem) A neural network has to be trained (using solved problems) in order to learn how to solve the problem (association) C \ BY: $ (LSI-FIB-UPC) Artificial Intelligence Architecture of KBS Term 2009/2010 19 / 24 Other methodologies Neural networks Notes input1 weight1 s weight2 f(inp1,...,wei1,...) input2 weight3 input3 C \ BY: $ (LSI-FIB-UPC) Combination Activation Artificial Intelligence Architecture of KBS Term 2009/2010 20 / 24 Other methodologies Neural networks Hidden Layers Output Layer SOLUTION EXAMPLES Input Layer Notes . C \ BY: $ (LSI-FIB-UPC) Artificial Intelligence Term 2009/2010 21 / 24 Architecture of KBS Other methodologies Intelligent Agents/Multiagent systems Notes It supposes a collective vision of intelligent systems (instead of monolithic) An intelligent agent solves only a simple task An agent: Obtains information from the environment (peception) Elaborates a decision based on its perceptions and state (reasoning) Performs an action (actuation) Agent C \ Sensors Decision Mechanism Actuators Perception Action Artificial Intelligence Architecture of KBS Environment BY: $ (LSI-FIB-UPC) State Term 2009/2010 22 / 24 Other methodologies Intelligent Agents/Multiagent systems Notes The global problem is solved by cooperation/coordination Techniques involved: organization, cooperation, coordination, negotiation, tasks distribution, communication, reasoning about others, ... C \ BY: $ (LSI-FIB-UPC) Artificial Intelligence Architecture of KBS Term 2009/2010 23 / 24 Other methodologies Intelligent Agents/Multiagent systems Notes Advantages: More flexible systems Reconfiguration/reorganization allow to solve other tasks =⇒ Agents act as reusable components Fault tolerance (an agent can be substituted by other) Distributed computing Related to: Grid computing/Cloud computing (Management of tasks and resources) Web services (No reasoning capabilities) C \ BY: $ (LSI-FIB-UPC) Artificial Intelligence Term 2009/2010 24 / 24