Presentation

F.A.Q.

What is a knowledge base ?
What is IKBS ?
Why a knowledge base ?
What is excellence in a knowledge base ?
Necessary means for achieving excellence in such a knowledge base ?
The reasons for remote collaboration ?
Methodology of the construction of a knowledge base

	Animated demonstration (in French)
	The Shockwave plug-in is required.
	Background of the project
	Telesystematics
	Seminar of numeric-symbolic group (in French)

What is a Knowledge base ?

A knowledge base is an application stocking the knowledge of an expert in a determined field. It incorporates a part of the knowledge possessed by a known specialist in a domain. This knowledge is multi-faceted, our knowledge bases deal with observable, observed and derived knowledge. Our knowledge base built with IKBS is composed of a descriptive model of the domain and a case base of comparative examples described by the same model (acquisition phase). Our aim is to derive new expert knowledge from the primary elements in the knowledge base, such as classifications or rules (discrimination and characterization of descriptions), and identifications (the result of naming new observations) which are as reliable as those of an expert.
A knowledge base is more general than an expert system and functions with the following elements :

A conceptual model of the field
Examples, cases, situations verified by experts (name validation)
Learning and case-based reasoning techniques to generate classifications (by generalization of specific cases) and identifications (by comparison of specific cases).
Hypertextual (for navigation) and multimedia techniques to facilitate access to the content to the greatest number of visitors.

Thus, a knowledge base is not a database nor an information system. The knowledge is more general than the data and can thus be exploited at differing levels (observable, observed and derived knowledge).

What is IKBS ?

IKBS (Iterative Knowledge Base System) is a software workshop developed at the IREMIA for constructing knowledge bases. It facilitates the management and communication and saving of knowledge in all the fields necessary to ultimate decision making. IKBS was conceived to manage knowledge in the field of environmental sciences (characterized by a greater complexity in the cases to be described). It can also be used for industrial applications (characterized by a greater quantity of less complex cases to be described). It is intended to apply experimental method to knowledge base creation (observation of facts, construction of hypotheses, testing and validation). This learning method, being one of inductive reasoning is better adapted to the life sciences such as biology and medicine.

IKBS proposes 'descriptive logics' (temporal and spacial composition, viewpoints, specializations, particularities, iterations, contextual conditions) enabling one to naturally model a domain in such a way as to approximate expert modeling. IKBS helps to adapt the model to the reality of the domain, not the contrary. For example the semantic layer of the descriptive logics is superimposed over the layer of the model of knowledge representation of the system. IKBS adopts an object oriented representation which is necessary to represent living beings in knowledge bases.

Thus, IKBS is a generator of knowledge bases that assists the user in the task of modeling, description, classification, identification, updating and dissemination of knowledge. Its originality lies in its capacity to iteratively manage evolving information by reconsidering and updating the descriptive model and cases.

Why a knowledge base ?

Expertise is a rare commodity which is insufficiently exploited and badly passed from generation to generation (the master-student relationship is becoming more and more rare). The knowledge base enables the user to do three things :

stocking : the depiction and conservation of knowledge
management: manipulate, compare, process knowledge
transmission : communicate and disseminate expertise

What is excellence in a knowledge base ?

Above all, a knowledge base must be an efficient (easily handled) and robust application yielding results that are correct from an expert's point of view.

To fulfill the requirements of robustness it must be: dependable, comprehensible, precise, exhaustive, coherent, redundant, ergonomic, noise tolerant and well adapted to requirements.

It is not a 'black box' : knowledge is explicitly represented and bad decisions are easily traceable.

The role of a knowledge base is to pin down and unearth expertise with the assistance of automatic learning techniques based on examples. These techniques of knowledge extraction don't rely on the traditional interview method of eliciting knowledge used in the construction of first generation expert systems.

Necessary means for achieving excellence in such a knowledge base ?

Rich language of knowledge representation to deal with the complexity of the domain through quality descriptions.
Iterative (incremental) aspect of application development to deal with the evolution of knowledge
Use of cognitive (structuring of knowledge), visual, pedagogic and multimedia means for intelligible vulgarization to assist future users of the application.
Availability over the Web.

The reasons for remote collaboration ?

Experience shows that team work facilitates the creation of more robust knowledge bases than when contributions are received from isolated specialists. The pooling of knowledge is beneficial in creating a shared thesaurus for vocabulary and illustrations.

The knowledge pooling determines the user's ultimate acceptance of the application because the team-work guarantees of quality of the knowledge base.

Methodology of the construction of a knowledge base :

Acquisition of a descriptive model. (background knowledge or observable data),
Acquisition of examples constitutive of observed data, assisted by a questionnaire,
Processing of the above-mentioned knowledge by differenciation (induction) and comparison (case-based reasoning),
Verification of derived knowledge,
Iteration over the definition of the descriptive model and update of older cases.

Concept innovation : first experience of construction of a multi-expert knowledge base with widely dispersed participants (Antillas, continental France, Runion Island).