Omnispective analysis and reasoning: an epistemic approach to scientific workflows

Abstract

This thesis presents the conceptualization, formulation, development and demonstration of Omnispective Analysis and Reasoning (OAR), an epistemic framework for managing intellectual concerns in scientific workflows. Although scientific workflows are extensively used to support the management of experimental and computational research, intellectual concerns are not adequately handled in current practice owing to the focus on low-level implementation details, limited context support, issues in developing shared semantics across disciplines and lack of support for verification and validation of the underlying science of the workflow. The management of intellectual concerns in scientific workflows can be improved by developing a framework for providing a layer of abstraction to lift focus from low-level implementation details, adding context as a workflow parameter, introducing localized ontologies and abstracting and mapping intellectual concerns in the research-domain to workflow specification and execution semantics. Following an examination of typical definitions of scientific workflow offered in literature, the Scientific Method is applied to develop an enhanced definition of a scientific workflow. This definition, which extends the scope of ordered analysis and investigation to a generic problem scenario, is utilized in the OAR framework. The design of OAR is modular like the Domain of Science Model (DoSM). The structure and working of OAR incorporate the evolving nature of science, hierarchy of conceptualization, omnispection, and the logical processes of analysis, reasoning and abstraction. These form the Foundation and Theory of OAR. Abstracting concerns in terms of unit knowledge entities (ukes) and groups of ukes (recipes), use of context to identify relation between recipes, the management of recipes in shelves, and the processes of concern refinement and context refinement constitute the Methodology. A comprehensive and simple example of the application of OAR to the abstraction, analysis, formulation and orchestration of a scientific workflow at different levels of granularity is provided by applying it to the problem of origami paper folding. The use of OAR in capturing the rationale of design decisions and mapping them to desired outcomes is demonstrated by applying OAR for contextualizing course design. Another example illustrates the use of OAR in the analysis, understanding and management of complex systems. Localized ontologies enable the exposure of side-effects and emergent behavior in large-scale systems due to the choice of any particular solution specification. These examples constitute a first step in building the Applications block of OAR. While OAR may be manually applied even to large-scale problems, it is expedient to avail of tool support. Soma - a simple and illustrative tool prototype is developed to indicate directions for a reference tool implementation. The thesis concludes with a consideration of ideas for future work. The contribution in this thesis corresponds to an instance of the DoSM for scientific workflow management. The OAR framework has great potential for further development as a well-formed Science of Workflows.