suiji

Machine learning has come into its own as a driver of technological innovation.  The field borrows heavily from the more mature, though still rapidly evolving, disciplines of statistics and optimization.  Perhaps as important, but less well touted, is the contribution of performance, which is essential for operation at scale.  The role of cluster-level parallelism, for example as availed by Hadoop, is well understood.  Opportunities for lower-level performance, such as SIMD parallelism, are not as frequently addressed.  Suiji provides both a forum, through which the more hardware-oriented aspects of performance can be explored, as well as open-source software, to implement and illlustrate many of these ideas.