Model Management Projects

Supervised by - Dr Gee Kin YEO
Associate Professor, Department of Information Systems, School of Computing, National University of Singapore

Model management is concerned with the representation and manipulation of models and aims to provide (automated) support for various phases of the modeling life cycle. Decision models are an important organizational information resource and sh ould be managed like stored data.

Right here in National University of Singapore, the following model management research topics are being pursued:

Visual decison modeling

Decision modeling in time dimension

An object-oriented approach to model integration

A knowledge-based approach to some multi-criteria modeling

Visual decision modeling is about using graphical representations to assist modeler and end-user in modeling. Decision modeling in time dimension covers the dynamic aspect of the modeling process, an aspect which many other existing modeling systems has neglected. Objected-oriented approach to model integration tries to enhance the power of existing models by integrating them to perform a greater function using an object-oriented approach. And lastly, as most of the real world problems involve many knowledge domains and at times, other than simply applying rules and formulas, human's intuition and perceptions are necessary in problem solving. By using artificial intelligence, this aspect of problem solving is addressed in the knowledge-based approach to some multi-criteria modeling.

Read on to find out more.

Visualization of Structured Modeling

Graphical or pictorial representations are very useful for modeler and end-user to understand complicated relations and structures of decision models and can be used intuitively to construct them. This is the reason that visualization of decision model on computers is becoming more and more important for modeler and end-user. When the executable language used to describe analytic models is a kind of visual language which consists of a set of visual units, which includes icons; menu-items; dialog-windows; lines; boxes; circles and text strings, modeling becomes visual modeling. Visual modeling allows the user to create and manipulate graphical objects (models) and perform actions on them that will be understood by the system. This will hence free the user from handling unintuitive command syntax of executable languages such as GAMS, AMPL, or SML used in a MS/OR model.

Visual programming is the use of graphic representations in the process of programming[Rathnam and Mannio, 1995]. Likewise, visual modeling is the use of graphical representations in the process of modeling. With visual modeling, non MS/OR specialists will definitely find it easier to model problems compared to using the conventional modeling methods. In this project, the possibility of using visual structured modeling to assist user in modeling is explored.

A Prototype - VMS/SM

Published Papers

Status - Closed
Done by
- HU Jian
- LOH Seow Yick
- HUANG Liling

see also Structured Modeling

Discrete Time Modeling

A modeling system deals with three aspects of a model: the conceptualization or abstraction of the model, the information or representation of the model, and the data specification of the model. In MS/OR modeling which very often involves management processes with dynamic change, time needs to be modeled along with model abstraction so that it can be mapped correctly into the representational or informational aspect of the model. The few attempts at these two aspects of time modeling have been mostly done in a conceptual and general way.

Many existing MS/OR modeling systems, except simulation systems, are only concerned with the static status of the real world problems. They provide very little direct guidance in the design of temporally oriented models. In modeling system, further development of effective modeling capabilities for time modeling critically depends on an adequate understanding of the temporal nature in MS/OR problems. Such an understanding would form the common basis for the design of a modeling language with time modeling capabilities which are useful for dynamic models.

Time modeling in MS/OR is a challenging area. In this project, time requirement in modeling is analysed and we try to identify the key concerns of time constructs in MS/OR models which we think are the common basis for the design of a modeling language with the time modeling capabilities useful for dynamic models. How the existing modeling languages deal with these time concerns is also reviewed. A temporal extension for SML to illustrate how this extension is applied to time models. The possibility of using temporal modeling framework for time query operations is also explored.

Published Papers

Status - Completed
Done by - Li Guoqiang

Model Integration via an Object Oriented Model Representation

Modeling and artificial intelligence are the core of advanced decision support system. A model representation is the core of modeling. It should be combined with artificial intelligence to enhance the power of models to support decision making. The challenge is how to access existing models, generate new models and integrate suitable models for a comprehensive problem. An effective internal model representation should therefore be considered. During the last decade, the major advances in model management have been in the area of model representation. The typical one is structured modeling. Most of these researches focus mainly on model definition, which is referred an external model representation in. The challenge in area of model manipulation is to construct efficient internal model representation in which existing models can be reused conveniently in decision support systems.

A model representation with logic and object, OOMR, is presented in this project. OOMR is an internal model representation with an object oriented architecture including aggregation structures and specialization structures. It establishes an inference mechanism among different model abstract levels: model type, model template and model instance, that form a model class hierarchy. OOMR couples the object oriented architecture with logical reference. It's internal model representation is used as a foundation to implement the intelligence model management in Decision Support Systems (DSS). The OOMR is discussed in three parts: model piece, model class and model framework.

The model piece is the basic element of OOMR. A group of similar information is abstracted to a model piece class. In terms of the model framework, model piece classes serve as the attributes of a model class. Model classes of different abstractions form a model class hierarchy. The relationships between classes are represented in logical predicates. The model piece classes, model classes and model class hierarchy form an object oriented architecture for the internal model representation. Then, based on the object oriented architecture, a model integration approach is presented. In this approach, visual interface of the structured modeling is used to the user view of the model integration. So model schema integration and process integration are linked through OOMR. Model class and genus classes builds semantic association with the model schema and genera. It contributes to overcome the difficulty of identifying synonyms and homonyms. The integration of model schemas will be implemented by joining genera.

Published Papers

Status - Completed
Done by - Wang ShiPing

see also Structured Modeling

A knowledge-based approach to some multi-criteria modeling

There are two types of problem according to Newell and Simon: structured and unstructured problem. Unstructured problems are normally too difficult to be modeled by conventional models as most conventional models cannot handle complex decisions in an unstructured problem. Our research is based on a cross-disciplinary approach that utilizes AI techniques and OR methods. With this combination, we hope that unstructured problems can be modeled more effectively. Generally speaking, AI emphasizes more on the qualitative aspects of problems whereas OR emphasizes more on the quantitative aspects of problems. Model management is concerned with the representation and manipulation of models and it aims to provide (automated ) support for various phases of modeling life cycle. The main features of a MMS are

  1. ability to handle models in the same way that DBMS handle data
  2. algorithmic-independence, and
  3. a knowledge-based structure.

In regards to the knowledge-based capabilities of MMS, it is reasonable to expect that the current surge of interest in Artificial Intelligence will lead to a linking of Expert Systems technology with modeling and model management. From this view point, the Expert System will perform functions that can assist the modeler in representing and manipulating models, and this is illustrated by using Bank Loan Credit Evaluation as a domain problem.

A bank loan credit evaluation can be a difficult problem as most of the decisions are based on the loan officer's intuitive and experience, his perspicacity and overall impression of the loan applicants. A bank provides loans to many types of borrowers with a variety of reasons. Loans therefore will vary in liquidity and risk. Furthermore, most of the assessments are made with incomplete and uncertain data.

The paper A DSS for Bank Credit Evaluation under Risk has been presented in the INFORMS Singapore International Meeting held in June 1995 ( INFORMS -- Institute for Operations Research and Management Science). Implementation is based on Allegro CL for Windows and GBB to support our system. Allergo CL for Windows is the most advanced dynamic object -- oriented programming system for PCs, and GBB is a General Blackboard Builder developed by University of Massachusetts. The Blackboard model has been proven to be popular for AI problems and it can support more sophisticated goal-directed control strategies. The system is composed of three main components :

  1. a dynamic processing system
  2. model base , and
  3. the message board

The dynamic processing system chooses a course of action on the basis of the state of the message board. The choice of the action represents the expert's process when making a decision. Model base is a collection of problem-solving models designed by the domain expert, each model in the model base is a specialist at solving certain aspects of the overall problem. It can be independently executed like knowledge sources as long as model assumptions are satisfied and input values are defined. The message board consists of the model action table, the decision action table, the applicant report card and a data base. The model action table is a set of rules which keep tracks of the knowledge that is activated. The decision action table and applicant record card are two seperate databases maintained by the message board to keep track of the decision criteria to be apply to the loan applicant for evaluation.

Published Papers

Status - Completed
Done by - Ding WenXuan

Last updated: November 11, 2001.