670 likes | 808 Views
Integrating case-specific experiences with general domain knowledge for intelligent information processing. Agnar Aamodt Norwegian University of Science and Technology (NTNU) Department of Computer and Information Science (IDI) Division of Intelligent Systems (DIS)
E N D
Integrating case-specific experiences with general domain knowledge for intelligent information processing. Agnar Aamodt Norwegian University of Science and Technology (NTNU) Department of Computer and Information Science (IDI) Division of Intelligent Systems (DIS) Artificial Intelligence and Learning Group (AIL)
The Issues • Case-based reasoning (CBR) intro - basic, and knowledge-intensive • Knowledge modeling and representation - Merging general and situation-specific knowledge • Reasoning: Knowledge-intensive case retrieval - Syntactic and explained matching • Development and testing tools - Knowledge modeling editor and case matching visualizer • Recent and ongoing research • Challenges ahead
Case-Based Reasoning • CBR is a technology for solving a new problem by retrieving a similar case whose solution is known, and adapting its solution. New Solution Past Solution New Problem Past Problem Case-1 General knowledge • It is also a technology for machine learning, by integrating a new case just solved into the exisiting case base.
Knowledge-Based Methods - Development History Heuristic rules (e.g.: MYCIN) Rule-based systems -> Rule-based reasoning
Knowledge-Based Methods - Development History Control Knowledge Heuristic rules Explicit Control Knowledge (e.g. NEOMYCIN) -> Meta-level reasoning
Knowledge-Based Methods - Development History Control Knowledge Heuristic rules Deep knowledge Deeper models, text book knowledge (e.g.: CASNET) -> Model-based reasoning
Knowledge-Based Methods - Development History Control Knowledge Specific Heuristic cases rules Deep knowledge From generalized to situation-specific knowledge (e.g. PROTOS, CHEF) -> Case-based reasoning
Knowledge-Based Methods - Development History Control Knowledge Specific Heuristic cases rules Deep knowledge Integrated systemes (e.g. SOAR, THEO, META-AQUA, CREEK) -> Architectures for intelligence
The Case-Based Reasoning Cycle Problem New RETRIEVE Case Learned Case Retrieved New Case Case Past Cases RETAIN General Knowledge REUSE Tested/ Repaired Case Solved REVISE Case Suggested Confirmed Solution Solution
The ”knowledge-intensiveness” scale of CBR MBR CREEK IBL/IBR • No explicit gen. knowledge • A lot of cases • A case is a data record • Simple case structures • Global similarity metric • No adaptation • Learning is simple storage • Substantial gen. knowledge • Not very many cases • A case is a user experience • Complex case structures • Sim. assessm. is an explanation • Knowledge-based aptation • Knowledge-based learning
What is this thing called Knowledge? Focus: A decision-making process: 2 important perspectives on knowledge: - its role - its frame of reference
Knowledge and cases So, why do CBR systems need general domain knowledge to become knowledge-intensive? Aren’t cases knowledge themselves? Cases are sometimes viewed as data/information, not knowledge. Cases are often viewed as shallow knowledge. Additional general knowledge deepens - and widens - the system’s knowledge. Cases are knowledge, but their knowledge content is enhanced by their features being defined and related within a body of general knowledge.
• Combines case-based and model-based reasoning, for problems in openandweak theory domains. • Input is problem solving context (e.g. goal) and problem features (e.g. a list of findings). Output is the bestplausible interpretation of the input within the context. • Knowledge types, used for reasoning are • a body of situation-specific knowledge, • i.e. a case memory of findings linked to • solutions, annotated with other relevant • information and knowledge • a body ofgeneral domain knowledge, as deep • relationships or heuristic rules The Creek1 Approach 1Case-based Reasoning through Extensive Explicit Knowledge
thing goal hsc hsc case hsc domain-object hsc hsc hsc diagnosis find-treatment find-fault hsc has-output described-in vehicle case#54 has-function hi van transportation hsc hsc has-status diagnostic-case solved tested-by car hp hd hp wheel test-procedure possible-status-of hp test-step has-electrical-status hp hp engine hi has-state hsc starter-motor-turns has-engine-status has-fault tested-by hsc fuel-system case-of electrical diagnostic-hypothesis -system engine-test N-DD-234567 hsc has-fault engine-turns hp car-fault hsc test-for battery-low subclass-of has-fault fuel-system-fault engine-fault hsc instance-of hsc battery hsc broken-carburettor-membrane subclass-of subclass-of has-fault electrical-fault hsc status-of part-of battery-fault observed-finding tested-by hsc - has subclass finding subclass-of turning-of hi - has-instance test-for hp -ignition-key - has-part starter-motor hd - has-descriptor Model-Based Reasoning MMo • MBR - in this context - is a technology for solving a new problem by explaining its solution within a multi-relational model of the target system. • MBR - in this context - involves the abductive steps of hypothesis generation and evaluation/selection, for which methods of plausible reasoning are applied.
thing goal hsc hsc case hsc domain-object hsc hsc hsc diagnosis find-treatment find-fault hsc has-output described-in vehicle case#54 has-function hi van transportation hsc hsc has-status diagnostic-case solved tested-by car hp hd hp wheel test-procedure possible-status-of hp test-step has-electrical-status hp hp engine hi has-state hsc starter-motor-turns has-engine-status has-fault tested-by hsc fuel-system case-of electrical diagnostic-hypothesis -system engine-test N-DD-234567 hsc has-fault engine-turns hp car-fault hsc test-for battery-low subclass-of has-fault fuel-system-fault engine-fault hsc instance-of hsc battery hsc broken-carburettor-membrane subclass-of subclass-of has-fault electrical-fault hsc status-of part-of battery-fault observed-finding tested-by hsc - has subclass finding subclass-of turning-of hi - has-instance test-for hp -ignition-key - has-part starter-motor hd - has-descriptor
t h i n g g e n e r i c c o n c e p t s g d o m a i n c o n c e p t s c a s e s c a s e c a s e c a s e 0 3 9 7 6 1 1 2 The Creek Architecture: Basic Knowledge Types l e n e r a 1Case-based Reasoning through Extensive Explicit Knowledge
Example: Knowledge-level modeling -> The Components of Expertise framework Knowledge engineering as modeling Problem solving as modeling
ACTIVATE EXPLAIN The Creek Explanation Engine • Goal • Goal - Appl. task accomplished - Appl. task is defined • Situation • Situation - Findings explained - Findings are listed - Constraints confirmed - Constraints are specified - Solution found - Solution asked for FOCUS
Main network inference approach: Plausibel Inheritance (Cohen and Loiselle, 1988) standard method extended method
water associated with used for bacterial epidemic drinking subclass of subclass of has solution caused by dirty water caused by clean water supply associated with bacterial infection epidemic case #3 bad hygiene Example of Plausible Inheritance Inheritance rules: Tt set of relation-types transferable through relation t (ct) set of relation-types transferable through concept c along t Ri set of relationships inherited to the initial concept Formalism: I = ((subclass of, causes), (subclass of, caused by), (subclass of, associated with), (subclass of, used for), (subclass of, has solution), (causes, causes), (caused by, caused by), (caused by, has solution) (associated with, associated with))
Example of Plausible Inheritance Initialization water associated with used for bacterial epidemic drinking subclass of subclass of has solution caused by dirty water caused by clean water supply associated with • = {t} bacterial infection epidemic case #3 bad hygiene Question: Find the extended frame for ’epidemic case #3’ or Check if ’clean-water-supply’ can be a solution to ’epidemic case #3’ or Find plausible solutions for ’epidemic case #3’ Ri = {}
First step Tsubclass of ={ causes, caused by, used for, has solution, associated with } ={ causes, caused by, used for, has solution, associated with } bacterial epidemic subclass of Tcaused by ={ caused by, has solution } caused by ={ caused by, has solution } bacterial infection epidemic case #3 I = ((subclass of, causes), (subclass of, caused by), (subclass of, associated with), (subclass of, used for), (subclass of, has solution), (causes, causes), (caused by, caused by), (caused by, has solution) (associated with, associated with)) Ri = {(‘epidemic case#3’, subclass of, ‘bacterial epidemic’), (‘epidemic case#3’, caused by, ‘bacterial infection’)}
(ct) = Tt I (ct-1) ={ associated with } ={ causes, caused by, used for, has solution, associated with } water associated with used for bacterial epidemic drinking subclass of subclass of has solution caused by dirty water caused by clean water supply associated with = T bacterial infection epidemic case #3 ={ caused by, has solution } bad hygiene I = ((subclass of, causes), (subclass of, caused by), (subclass of, associated with), (subclass of, used for), (subclass of, has solution), (causes, causes), (caused by, caused by), (caused by, has solution) (associated with, associated with)) Ri = {(‘epidemic case#3’, subclass of, ‘bacterial epidemic’), (‘epidemic case#3’, caused by, ‘bacterial infection’), (‘bacterial epidemic’, associated with, ‘dirty water’)}
={ caused by, has solution } ={ caused by, has solution, associated with } ={ causes, caused by, used for, has solution, associated with } water associated with used for bacterial epidemic drinking subclass of subclass of has solution = caused by dirty water caused by clean water supply associated with = T bacterial infection epidemic case #3 bad hygiene ={ caused by, has solution } ={ associated with } I = ((subclass of, causes), (subclass of, caused by), (subclass of, associated with), (subclass of, used for), (subclass of, has solution), (causes, causes), (caused by, caused by), (caused by, has solution) (associated with, associated with)) Ri = {(‘epidemic case#3’, subclass of, ‘bacterial epidemic’), (‘epidemic case#3 ’, caused by, ‘bacterial infection’), (‘bacterial epidemic’, associated with, ‘dirty water’), (‘bacterial infection’, caused by, ‘dirty water’), (‘dirty water’, has solution, ‘clean water supply’), (‘dirty water’, associated with, ‘bad hygiene’)}
CREEK Retrieve • - context focusing by spreading activation in the semantic network, followed by - index retrieval of possible cases, followed by - explanation-driven selection of best match Reuse • - attempts to copy solution from matched case - explanation-driven adaptation, by combining explanantion of retrieved case with general domain model Revise • - user evaluates and gives feedback - case status info kept and used in case selection and reuse Retain • - attempts to merge the two cases - stores relevant findings, successful and failed solutions, and their explanations - updating the strength of indexes
Ongoing projects (2003- • Knowledge-based learning support (IKT & Læring, NTNU) • AmbieSense – Context-sensitive info for mobile users (EU) • COST 282 Action – Knowledge exploration in sci & tech (EU) • Explanation of gene-gene relationships (Bioinformatikk, NTNU) • Avoiding unwanted events in drilling (Internal) • AI systems in the working environment (Internal) • From text to structured cases - data/text mining (Internal) • Conversational CBR for software component retrieval (Internal)
Avoiding unwanted events in drilling With Pål Skalle, Martha Dørun Jære, Inge Valaas, Elise Bakke, Tore Brede The Goal Predicting possible unwanted events during a drilling operation, in order to steer away from them and hence avoid them. Overview Temporal case-based reasoning - based on temporal intervals - is used to represent cases. The CBR process supervises the drilling prameters, and gives a warning whenever a situation occurs that reminds of a past situation that failed. If so, CBR and MBR are used to suggest how to avoid the potential unwanted event.
Component Retrieval Using Conversational Case-Based Reasoning With Mingyang Gu and Torgeir Dingsøyr The Goal A method that will combine knowledge-intensive CBR with the Conversational CBR method, applied to software component retrieval from component libraries. Overview Component retrieval, about how to locate and identify appropriate components, is one of the major problems in the component reuse. And it becomes more critical as more reusable components come from component markets instead of from an in-house component library, and the number of available components is dramatically increasing. In CCRM, components are represented as cases, a knowledge-intensive case-based reasoning (CBR) method is adopted to explore semantic similarity between users’ query and stored components, and conversational case-based reasoning (CCBR) technology is selected to acquire users’ requirements interactively and incrementally