A Multi-Layer Architecture for Knowledge-Based System Synthesis

This paper deenes a clean multi-layer architecture for the design of knowledge-based synthesis systems within the logic programming paradigm. The intent is to provide an integrated logical framework for modeling the diierent kinds of knowledge involved during synthesis processes, a workbench of inference-based generic methods for the constructive solution of knowledge acquisition and synthesis tasks. As major advantages, evolutionary synthesis processes, the re-usability of knowledge chunks, and multiple views are facilitated and correctness criteria are provided. The approach depends upon the deenition of hierarchically organized abstraction layers: domain, domain abstractions, and nally, distributed knowledge sources. Each layer is partitioned into data type, declarative, and operational speciications. For the interaction between heterogenous representation schemes and abstract representations the concept of logical viewpoints is introduced. 1 Motivation \A programming environment suitable for prototyping would support a design methodology that emphasizes creativity, experimentation, learning and evolution." ((Flo84]), p. 13) Formal correctness and conformance with user needs are criteria to assess the quality of software. The \proofs-as-programs" paradigm, ie. assuming declarative , rst order domain speciications, proving requirement speciications in a constructive manner by (inductive) proofs and elaborating algorithms and programs from proofs, has been proposed in the area of logic programming as a theoretical basis for correct program synthesis. The deductive approach covers only parts of automated development processes due to the inherent complexity (cf. Low93]) and the incompleteness of uniform proof procedures. One of the major problems is modeling in advance domain and requirements in a non-ambiguous, complete, correct, and adequate way. Thus, speciications are subject to continual changes. Experiences gathered so far from automatically solved complex algorithmic problems 2 suggest that computer-assisted discovery of solutions ? This paper is a shortened version of Eus95]. 2 Eg. the computer-assisted proof of a optimal depth lower bound for nine-input sorting networks Par91]. for non-trivial problems involves-besides the guidance by abstract principles-comprehensive domain speciic and strategic knowledge. The intent of the SEAMLESS approach is to enhance the derivation of correct , eeciently executable speciications from declarative speciications by providing an integrated knowledge based synthesis environment which supportstive system synthesis is perceived as a non-linear, incremental knowledge acquisition process. Thus, we stress the incremental nature of software development. This contrasts with the traditional approaches, usually based on deductive rst order logic or metalogical frameworks. While these are intended to be generally applicable, the domain modeling and the derivation of methods remain ad-hoc processes, which have …