Custom domain-specific architectures have great promise for achieving energy-efficient flexible designs for a suite of applications. For practical deployment, however, there is a great need to develop smart algorithms for mapping applications of interest onto these architectures. The main thrust of this research is to discover novel mapping algorithms by making use of human intuition and ability to recognize patterns and opportunities even in complex problems. Participants are presented with successively more difficult mapping problems in a game environment, and the vast dataset of participants' moves is analyzed to recognize common patterns used by successful game participants. The insights gained from strategic moves humans make while solving problems based on their visual intuition and experience can be used to discover new mapping approaches that are beyond what can be conceived with traditional algorithms. For broad scope and impact, this research places particular attention on multiple architectural designs, highly constrained (and thus very difficult) mapping problems, and engaging a broad community of contributors through crowdsourcing.
Fast and effective mapping techniques, and a methodology for developing new such techniques as needed has the potential to inspire architectural innovation and result in production of devices that are faster, cheaper, and smarter. Results from this research can advance next generation portable/wearable devices critical to health, safety and security, personal multimedia, and aerospace.
