On Tue, Mar 21, 2017 at 11:35:06AM +0800, 孟憲妤 wrote: > Hi , > > Since last time we discussed about memory partition of SGD, I did some > literature review on single-machine and multi-machine parallel > machine-learning approaches. It seems like that GPU-based learning is the > dominent form of parallelism among single machine approaches. I found this > allreduce > <https://www.microsoft.com/en-us/research/wp-content/uploads/2016/02/IS140694.pdf> > approach > is quite interesting. It implements a data-parallel distributed SGD and > allows both parameter averaging, gradient aggregation, and iteration. I'm > interested in particating in SGD project. Do you consider implementing this > scheme appropriate in a GSoC project ?
Hi Hsienyu, There's no good abstraction for GPUs in mlpack at this time, so that might be difficult to put inside of mlpack. Ideally mlpack should provide a clean and consistent interface, and including GPU code probably would break that. But, there is a "secret project" in the works for a GPU-based Armadillo library, so that may solve these issues. :) (However, it may be some time until that is ready; it certainly won't be available in time for Summer of Code.) Thanks, Ryan -- Ryan Curtin | "Happy premise #2: There is no giant foot trying r...@ratml.org | to squash me." - Kit Ramsey _______________________________________________ mlpack mailing list mlpack@lists.mlpack.org http://knife.lugatgt.org/cgi-bin/mailman/listinfo/mlpack
