by Ben ChandlerDepartment of Cognitive and Neural Systems, Boston University
The push towards exascale computation has opened a number of fundamental architectural issues. Energy loss to communication, fault tolerance, massive parallelism, and locality are all areas of active research. For a restricted, but interesting, set of algorithms, biology offers architectural insight that can help us avoid these scaling problems in practice.
Cog ex Machina (Cog), a platform jointly developed by HP Labs and Boston University, is an attempt to leverage this biological insight for large-scale computation. Cog exposes a high-level language that allows rapid development of highly parallel algorithms. The framework transparently compiles the high-level representation to a set of GPU kernels and executes the full algorithm interactively across a cluster.
This talk will present an overview of the joint HP/BU approach and several intermediate research results.
Ben Chandler is a doctoral student in the Department of Cognitive and Neural Systems at Boston University and ACES associate at the university’s Center for Computational Science. His research interests include large-scale simulation, computational vision, and neuromorphic computing. Chandler received a BS in cognitive science from Carnegie Mellon University. You can contact him at email@example.com.