Efficient Memoization Strategies for Object Recognition with a Multi-Core Architecture

Architectural research is increasingly driven by application needs in different domains. Research and development efforts must systematically identify the prominent properties, especially the amount of inherent parallelism and memory access/structural patterns for a wide range of emerging applications. One such application domain is recognition. Machine vision, object/scene understanding, and natural language processing belong to this domain. Recognition systems solve what is called the inverse problem. Whether it is recognizing a target from SAR or image data, recognizing speech, understanding human language, identifying a person, or tracking a computer attack, the problem centers around recognizing an event, action, or object of interest in the presence of uncertainty. Highly data-parallel preprocessing along with multigranularity task and data-parallel learning and inference are characteristic of machine recognition systems.