Prime Offeror: Georgia State University (Georgia State University Research Foundation)

Subcontractor: Georgia Institute of Technology (Georgia Tech Research Corporation)

Technical Contact:

Administrative Contact:

Albertha Barrett

Director, Research and Sponsored Projects

University Plaza

Atlanta, GA 30303-3083

(404) 651-4350

Fax: (404) 651-4436

Type of Business: Other educational

Duration of effort: 27 months

As the Department of Defense becomes more reliant on computing technology to support all of its mission-critical and support functions, the risk grows that a future legacy of monolithic software applications and systems will be increasingly difficult to modify and adapt to rapidly changing needs. Computer systems are more complex than ever even as research and technology development efforts strive to make them more accessible and adaptable to a growing and increasingly diverse set of end-users. While mission-critical applications warrant the need for training and special skills, computing technology is pervading so many everyday activitites in the military, administration, business, and even domestic spheres that people with widely differing cultural and educational backgrounds, work roles and authorities, and physical and cognitive capabilities are called on to interact with computer-based systems through special-purpose equipment, general-purpose computer peripherals, and hand-held and wearable devices.

Such trends are accelerating the need for responsive, flexible and adaptable software without sacrificing the reliability and integrity that we expect from carefully engineered systems. We are proposing to develop and integrate technologies for meso-adaptation, a form of system adaptation intermediate in scope, timescale and difficulty between the traditional goals of enhancement and maintenance through global, assured but unresponsive software engineering processes, and the more recent trend in user customization and automated adaptation, which are very rapid, but narrow in scope and uncontrollable at the system level. Meso-adaptation works at the level of user conceptual models, reconfigurable and replaceable interfaces and policies, and reconfigurable components. It requires some skill and training from a change administrator but that role more closely resembles operations management and maintenance than engineering design or end-user fine-tuning. Recent research by the investigators and others in the recently completed EDCS effort as well as ongoing work will be adapted for meso-adaptation. In addition, new techniques will be developed for analysis of conceptual model variance formal assessment of adaptability. The project will result in a software technology for supporting meso-adaptation, MesoMorph, which will provide adaptability and similarity gauges for change administrators.

MesoMorph addresses the design, coordination and validation of adaptations, with an emphasis on their design:

• Design of adaptations is supported at the conceptual, specification and implementation levels. The ontology representation and tools address the conceptual modeling and evaluation of adaptations in the context of an existing application ontology. The specification of interface portals addresses the design and evaluation of external interfaces at a user level. And the specification of adpation wrappers (HAS Beans and IS Beans) addresses the detailed advertising of component ontology and contextual assumptions.
• Coordination of adaptations is supported through the use of cohesion and compatibility gauges, which facilitate the incorporation of features that preserve system integrity.
• Validation is addressed through the constraints on composition imposed by the wrapper architecture and by the gauges that support the change administrator’s decisions to incorporate or assemble components. These forms of validation are all formative in that they are conducted as an integral part of the synthesis of adaptations and not performed during subsequent review or testing activities as are "summative" validation processes (although summative evaluation is also certainly necessary for time-critical systems).

To avoid a proliferation of new human-readable language, we intend to embed the MesoMorph ontology representation WorldView in UML as a set of stereotypes and to provide an XML-based interchange format for it.

Task ON-1 will result in a review of ontology representations either as a short stand-alone technical report or forming part a larger report or paper.

A WorldView ontology is a graph-based structure consisting of nodes and edges that denote real-world concepts and associations and constraints affecting them. Since the beginnings of modern graph theory, mathematicians and social scientists have proposed analyses of graph centralization, cohesion and sub-structure. We are developing analysis strategies for ontologies and data models that incorporate such analyses.

This task will define centrality, density and cohesion metrics to identify those nodes and subgraphs that are in some sense central to the graph or most closely related to each other. We are surveying research areas broadly to identify the most appropriate metrics and analyses for ontology. The MesoMorph metrics will likely be based on one or more of the following:

• Measures of centrality obtained by computing average or maximum distance from other concepts. Closeness centrality has provided extraordinary analyses of traditional architectures and city plans in the "space syntax" paradigm [HIL84, HIL96]. In space syntax analyses, a building or city is represented as a graph, with nodes denoting interior spaces or straight-line segments of roads, and edges denoting permeability or adjacency. Centrality tends to correlate markedly with flow and occupancy, but not necessarily with geographical centrality or official significance. Thus, an emergent structural metric with no a priori planning significance provides a more reliable index of economic and social factors than political or geographical intuitions. For similar reasons, we anticipate structural metrics to be more useful in identifying conceptual and contextual similarities and points of compatibility than a designer’s or user’s intuitions about the system.
• Measures of prestige or "betweenness" obtained by computing the size of a node’s neighborhood or the number of geodesics (shortest paths between other pairs of nodes) on which the node in question falls. These metrics are fundamental measures of centrality in social network analysis [SCO91, WAS94], where nodes denote people or organizations and edges represent affinities or similarities between them. Prestige measures capture the intuition that important concepts are those with many associations. Betweenness measures capture the intuition that important concepts lie between many others.
• Measures of cohesion and coupling from traditional software metrics [BIE98] and discrete mathematical representations of conceptual difference derived from concept theory and statistical cluster analysis [MIR96]. All these analyses produce partitions, overlapping substructures, or a lattice of sets that represent the clumping or separability of parts of the graph. While they have been applied to architectural components and terms extracted from design documentation, they should also apply satisfactorily to ontologies.
• Measures of similarity or distance derived from cladistics [KIT98], in which biological morphometry is used in conjunction with hierarchical cluster analysis to establish lineages and evolutionary distance between species. These techniques have already been applied in software composition and reuse [AST97], but with the goal of constructing domain-specific component libraries.
• Graph-theoretic measures of substructure. These include the presence of pseudo-cliques or clans, the identification of relative points of articulation in the graph (e.g. using the k-cut criterion), and the identification of local centers (e.g. in facilities planning). The computational complexity of most of these analyses may preclude their interactive use on complex ontologies. But they can be applied to local substructures identified through other means.
• Qualitative indices of structural matching and mismatching derived from the branch of cognitive science that studies conceptual models, especially in science and education. Thagard [THA92] has argued that competing scientific worldviews (or a comparison of an earlier and later more elaborated model) differ in two ways: one difference is the presence or absence of concept subgraphs. Generally, later models admit more special cases or refinements. The other difference is the presence of structural mismatches that affect "has a" and "is a" associations. Similar, but more complex "Pathfinder" models have also been used in information science and the study of user models.

Although our emphasis is not the improved design of user interfaces at the articulation level, but rather the use of interface data to determine system ontologies, we can measure the conceptual coverage of portals and the coverage of series of portals in a specified activity scenario (see Task IF-5). This information can be used to streamline the dialog structure or broaden it so that closely affiliated but currently obscured information is revealed to the user when performing a critical task.

Task ON-2 will produce a review of metrics and their viability as the basis for concept gauges. The review may take the form of a stand-alone technical report or the body of a larger report or paper.

The MesoMorph ontology includes not only a representation but also the principles with which an ontology may be recovered from external interfaces, and in particular the specification of a canonical form for ontology comparisons and compatibility assessment. WorldView will be based on a practical semantic modeling framework such as NIAM [VER82] and its derivation principles be an extension of data normalization. NIAM and similar binary relationship models provide mechanisms for specifying quite rich domain-specific constraints without recourse to a separate annotation language such as UML’s Object Constraint Language.

Task ON-3 will produce a description of the principles underlying the MesoMorph ontology representation together with examples of external interfaces and their (manual) translation into ontologies. This description will take the form either of a stand-alone technical report or parts of a longer paper on MesoMorph.

The ontology gauges produced in task ON-4 are prototype implementations of analyses and metrics derived from existing analyses studied in task ON-2. The metrics and analyses themselves will be specified either in short stand-alone technical report, pages linked from those documenting the ontology representation (see ON-5 below) and/or sections of peer-reviewed papers on MesoMorph.

Initial feasibility prototypes will be implemented in a mathematics environment, such as Maple V, MathCad or SPSS. Gauges will be described in parts of peer-reviewed papers and in user documentation.

Whereas ON-3 elaborates the representational principles behind WorldView, ON-5 defines it precisely and provides illustrative examples for external users. (ON-4 is a research task, whereas ON-5 is an advanced technology packaging task.) During this task, we will specify abstract and concrete syntaxes for WorldView, the latter including a UML embedding or mapping that permits the some analysis of WorldView descriptions using industry-standard UML-based tools and an XML document type definition (DTD). We will construct examples taken from our ongoing case studies (see Section 5.4) as well as small illustrative examples from other domains.

The resulting specifications will take the form of specification and illustration pages at the project web site and either stand-alone specification documents or parts of larger documents on WorldView and MesoMorph.

WorldViewer will either be a WorldView XML viewer, probably implemented in Java, or a web page generator probably implemented in a portable scripting language. In either case, code will be integrated along with relevant gauges into the MesoMorph suite and made available together with examples on the project web site.

The ontology gauges produced in task ON-7 are software implementations of the ON-4 prototypes. They will be loosely coupled software components probably written in Java or a portable scripting language. Code will be made available through the project web site, including installation instructions, API specifications (in Javadoc in the case of gauges written in Java) and short user instructions.

• The use scenarios will follow the episodic goal-directed representation of ScenIC [POT99], and will concentrate on the elaboration of key scenarios that are identified by considering individual obstacles and combinations of them.
• User capabilities will not involve detailed user modeling or sensori-motor functioning (e.g. at the keystroke level [CAR83]). but rather the attribution of discrete or ordinal-scale capabilities to users when situational factors are fixed. These will include, for example, reading ability, handedness, and boolean task capabilities. These properties will be attributed to ScenIC actors, which currently are atomic named elements.

Task HF-1 will produce a description of the principles underlying MesoMorph context representations, together with examples of human factors, activity scenarios, and their (manual) translation into contextual abstractions. This description will take the form either of a stand-alone technical report or parts of a longer paper on MesoMorph

To avoid inventing another language, we will extend (and restrict) ScenIC and its XML representation SCML as supported by ScenIC View. To avoid confusion, because the goals of MesoMorph are different from the general software-engineering process support of ScenIC and MORALE, we refer to this enhanced representation as HASL.

In HF-2, we will define HASL and describe it like the other XML-based embeddings of MesoMorph representations by specification and illustration pages at the project web site and either stand-alone specification documents or parts of larger documents on MesoMorph.

HF-3: Develop HASL viewer

We will enhance ScenIC View by implementing standard views and reports of contextual information. These enhancements may be implemented on either open-source XML processing software or simple HASL-to-HTML filters. We anticipate being able to reuse most of the language-independent WorldViewer and PortL code as well as first-generation ScenIC code. For details of the deliverables, compare task ON-6.

The HASL metrics will include qualitative measures of scenario vulnerability to breakdowns, and rough quantitative indexes of feature versus work-around benefits (in terms of categories and degrees of goal achievement) and risks (in terms of likelihood and consequence of obstacles). We will integrate this work with ongoing research into feature use and feature-unavailability workarounds involving Dr. Potts, and Dr. A. Anton of North Carolina State University. We will also incorporate Anton’s similarity metrics for scenarios [ANT00].

The gauges will be prototyped like the ontology gauges, probably using a combination of mathematical packages (e.g. Maple V) for graph-theoretic path algorithms, and spreadsheets or statistical packages (e.g. SPSS) for plotting and computing conditional probabilities. We will describe the context metrics and gauges in parts of peer-reviewed papers and in user documentation accompanying their implementation.

The context gauges produced in task HF-5 are software implementations of the HF-4 prototypes. They will be loosely coupled software components probably written in Java or a portable scripting language. Code will be made available through the project web site, including installation instructions, API specifications (in Javadoc in the case of gauges written in Java) and short user instructions.

Task RM-1 will involve a short review of published architecture definition languages to identify those abstractions most suitable for interface-to-ontology and interface-to-context mappings. Part of this analysis will consist of an comparison of abstractions against the models produced by MORPH’s static analysis. We do not intend to adopt a non-executable design language, but will instead evaluate existing APIs and architectural interfaces to executable components. We anticipate having to define adaptability layers on top of the selected architectural abstraction(s) but using their vocabulary and interfaces.

We will describe the result of task RM-1 in a short survey report or a section in a longer report or peer-reviewed paper about MesoMorph.

The IS Beans interface is a definition of an adaptability architecture for interface-to-ontology mappings. It will be defined in terms of transformation types that are the inverse of TransPortal mappings, but since the goal is to abstract ontology features for future adaptability, the TransPortal and IS Beans abstraction instances are unlikely to be similar in most cases of system adaptation. We will keep closely to the Java Beans and EJB standards and guidelines in formulating IS Beans, and incorporate ontology and interface features directly into the component attributes.

The IS Beans specification will be published as a Javadoc API on the project website and will be accompanied with succinct design documentation containing rationale and illustrations.

This task is comparable to RM-2 with the substitution of HAS Beans for IS Beans and HASL-to-PortL mappings in place of WorldView-to-PortL mappings. Because the task dependencies for RM-3 are more complex and RM-2 and RM-3 can be performed independently, we consider them as separate work-breakdown items. For more details, however, see RM-2.

We will develop a migration engine, Morph++, that enhances the MORPH system [MOO98] and incorporates dynamic analyses in addition to MORPH’s existing static analyses, gauges to indicate the "distance" of a code abstraction from user input or output, and compatibility of a composition by measuring the coverage of IS Beans or HAS Beans against interface components. Our ideal of the MesoMorph process is to bring the interface coverage and composition coverage metrics into alignment so that the adaptability interfaces act as brokers between the ontology or context on one side and the interface code and user-visible interface components on the other. When both compatibility gauges are correlated across much of a system, there is evidence that the domain ontology and contextual factors are mapped to interface behavior and implementation. Many factors will disrupt this ideal picture in practice, but we intend to make the gauges easy to view together, compare and correlate, possibly with the aid of empirically determined normalization or scaling of values.

In addition, Morph++ will generate code wrappers for IS Beans and HAS Beans instead of platform-specific GUI code interfaces (as is the case with MORPH). The target interface abstractions can then be plugged into these Bean interfaces provided that they comply with the same standard.

Morph++ will be an extension of MORPH, and will therefore probably be implemented in C. The gauges may be stand-alone analysis modules or plug-in modules added to MORPH itself. They will be distributed as with the other major tool components through the project web site, along with installation instructions, illustrations, and design documentation. The migration engine and gauges will be described in a stand-alone report or sections of a peer-reviewed paper on MesoMorph.

We will perform the following two case studies to validate MesoMorph against its evaluation criteria. It is possible that an even better case study may emerge before starting these, so one or both could be replaced. The case studies will, however, be of comparable complexity. The output of both tasks will be reports or peer-reviewed papers and user demonstrations probably in the form of video recordings. MesoMorph-developed software will be available on the project website, although the back-end intervention systems, which are developed by others, may not be available.

For our first case study, we intend to exploit Dr. Moore’s existing affiliations with Emory University Hospital and the Center for Rehabilitation Technology to demonstrate aspects of interface migration when the target interface takes the form of neural implants (see Section 5.2.4. above). The context will be provided probably by the Broadband Telecommunications Center "Aware Home" currently under construction on Georgia Tech campus. This building is an authentic family dwelling, in addition to being a laboratory for ubiquitous computing technology. It includes a battery of advanced sensing and interaction appliances for computer-mediated performance of everyday interventions such as cooking, domestic chores, recreational viewing and listening, and so on. Thus the case study scenario involves demonstrating severely disabled individual’s control of his or her everyday environment.

Although this case study does not directly involve a defense application, there are clear implications for the separation of functionality from non-standard interfaces. Neural implants provide one of the most testing target interface technologies one can imagine today, and so we are confident that the results of this case study can be extrapolated to less exotic interface adaptation efforts.

The second case study will extend the ongoing dissertation research of R. Hobbs under the direction of Dr. Potts and under the auspices of the Army Research Laboratory into mechanisms for driving what-if simulations for the purpose of evaluating technology-induced changes in battle planning doctrine. Hobbs’s approach is to encode the narrative and the presentation of what-if scenarios in a representation developed from ScenIC. The software drivers for the back-end simulation program are "StoryBeans", and in the terminology of MesoMorph they consitute an amalgam of IS Beans (where the ontology is the doctrine and materiel simulated) and HAS Beans (where the context is the simulated environment and the simulated activities of participants). The intention is to enhance macro-adaptation (reprogramming the system over the course of months) and micro-adaptation (tweaking parameters during a run) with meso-adaptation (scenario-based planning of the next day’s exercises after the daily "hot-wash" debriefing session).

Feasibility studies of the principle have already been conducted at Ft. Knox, where the back-end is ModSAF. A proof-of-concept prototype implementation is about to begin by developing simulation-independent StoryBeans for a Java-based open-source battle simulation game. We will continue and extend this effort in MesoMorph by developing a more principled distinction between scenario-driver IS Beans and HAS Beans. We do not envisage being able to deliver robust software to a mission-critical simulation facility during the schedule of the project, but we will undertake detailed feasibility studies for translating the adaptation wrappers constructed in this case study to a simulation back-end used in real wargaming.

The results of the end-of-semester review will be incorporated into an interim report that will be provided to DARPA and also posted to the web. The schedule of research tasks will also be tracked and updated in this report. Other reports will be provided to DARPA as requested. The PI's expect to attend DASADA PI meetings as well, where the MesoMorph research results will be shared with the rest of the DASADA community.