| Summary: | This paper proposes a new design pattern, named <italic>Scoreboard</italic>, dedicated for applications solving complex, multi-stage, non-deterministic problems. The pattern provides a computational framework for the design and implementation of systems that integrate a large number of diverse specialized modules that may vary in accuracy, solution level, and modality. The <italic>Scoreboard</italic> is an extension of <italic>Blackboard</italic> design pattern and comes under behavioral type. The pattern allows for an integration of multimodal results, employing early, and/or late fusion paradigms. Additionally, it provides a framework for the evaluation of the modules, dealing with inconsistency and low accuracy. In this paper, the <italic>Scoreboard</italic> design pattern is described with the standard meta-data model, followed by a sample implementation. This paper also provides the evaluation results based on experiments and a case study. The evaluation results confirmed the robustness, modularization, ease of integration, efficiency, and adaptability of the solutions with the <italic>Scoreboard</italic> pattern in comparison with the <italic>Blackboard</italic> pattern and “no pattern” condition. This paper provides also a case study of <italic>Scoreboard</italic> application in an integration of emotion recognition results. There are certain complex problems in modern software engineering which require multi-stage, multi-party, multi-modal solutions, and non-deterministic control strategies. Among those are natural language processing, image processing, and emotion recognition, to name just a few. The proposed <italic>Scoreboard</italic> pattern might be used in the software addressing the problems, especially in research systems that explore large solution spaces and require runtime decisions on execution order.
|
|---|