SC17 Denver, CO

P88: PetaVision Neural Simulation Toolbox on Intel KNLs


Authors: Boram Yoon (Los Alamos National Laboratory), Pete Schultz (Los Alamos National Laboratory, New Mexico Consortium), Garrett Kenyon (Los Alamos National Laboratory, New Mexico Consortium)

Abstract: We implemented a large-scale neuromorphic algorithm called the Sparse Prediction Machine (SPM), on the Los Alamos Trinity supercomputer. Specifically, we used PetaVision, an open source high-performance neural simulation toolbox, to implement a 4-layer SPM applied to ImageNet video. Various optimization techniques were applied to efficiently utilize up to 8192 KNL nodes. The goal of the SPM is to predict future states of a system from a sequence of previous states, or in the case of video, to predict a subsequent frame from previous frames. In our training, the SPM was able to predict the 8th frame from the preceding 7 frames, including successful separation of foreground and background motion.
Award: Best Poster Finalist (BP): no

Poster: pdf
Two-page extended abstract: pdf


Poster Index