high performance computing on graphics processing units: hgpu.org

Pattern Recognition (PR) tasks are natural candidates for embedded systems, since they usually interact with humans and other complex processes in the real world. Often regarded as the part of Artificial Intelligence (AI) closer to perception, a typical PR application reacts to external events that the system perceives through physical sensors or input devices and produces a response using actuators or information display subsystems. Being usually far from trivial, very demanding from the computational point of view, and requiring a fast reaction time, PR algorithms constitute a real challenge to the embedded system designer. In this talk, some of the main application domains and optimization approaches proposed to deal with these relevant issues, along with many open problems and paths to improvement, are presented.