Attribute-Driven Design of Incremental Learning Component of a Ubiquitous Multimodal Multimedia Computing System

Abstract

System design using attribute-driven design (ADD) means that system requirements, including functional and quality requirements and constraints, are considered as drivers in the design process that yields the system's conceptual software architecture. The output architecture satisfies not only that the functional requirements but also the important qualities the system must possess. In ADD, the secondary qualities are satisfied within the constraints of achieving the most important ones. In this paper, we detail the design of our system's machine learning (ML) component using ADD. Tactics and primitives to achieve system qualities (i.e. performance, security, availability, modifiability, and usability) are essayed in this paper. The ML component of our system is responsible for (1) determining the appropriate media and modalities based on user context, (2) finding the replacement to a failed/missing device or modality, and (3) providing the context suitability of newly-added media or modality. The ML component's knowledge acquisition is incremental; it keeps its previously-earned knowledge in its knowledge database (KD) and appends newly-acquired ones onto it. The ML component makes the system intelligent, adaptive and fault-tolerant. This work on ML-based media and modality selection is our original contribution to the domain of intelligent pervasive human-machine interface