LSP 156, Low Power Embedded Analytics: FY15 Line Supported Information, Computation, and Exploitation Program
MIT Lincoln Laboratory Lexington United States
Pagination or Media Count:
This report covers the second year of the low-power embedded analytics project, a three-yearuniversity collaboration between Lincoln Laboratory and Professor Arvinds group at the MIT ComputerScience and AI Laboratory CSAIL. The goal of the project is to design and prototype a novelarchitecture that has wide potential applicability to important applications ranging from back-office bigdataanalytics to fieldable hot-spot systems providing storage-processing-communication services for offgridsensors. Speed and power efficiency are the key metrics.Current state-of-the art approaches for big-data aim toward scaling out to many computers to meetprocessing, storage capacity, and access bandwidth requirements. Data is distributed over manycomputers, and complex processing is decomposed into tasks that operate on localized data andaggregated back together. With an emphasis on scalability and cross-platform portability, applications arewritten in high-level languages such as Java. New systems and new algorithms can be put togetherquickly, but not optimal in terms of performance and power efficiency.