You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@horn.apache.org by "Edward J. Yoon" <ed...@apache.org> on 2015/09/23 04:17:05 UTC
[DISCUSS] Parallelization of Stochastic Gradient Descent
Hi forks,
Instead of row or column-wise partitioning strategy for parallel SGD,
I'm thinking about using of Pregel-style BSP model that makes it
possible to partition and process billion-vertex graphs in efficient
way.
But I'm really not sure whether it's fit for every-types of neural
network topologies. Can we think about this together?
--
Best Regards, Edward J. Yoon