systemml-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Matthias Boehm <mboe...@gmail.com>
Subject Re: [Discuss] GSoC 2018
Date Sat, 27 Jan 2018 22:23:38 GMT
yes, but we as mentors will help out as needed, especially on the initial
language and compiler integration. Furthermore, it's a rather scalable
project in the sense that it could also accommodate multiple students. For
example, the different backends are fairly independent sub-projects.

Regards,
Matthias

On Sat, Jan 27, 2018 at 1:57 PM, Nakul Jindal <nakul02@gmail.com> wrote:

> This is awesome!
> I am guessing the goal is to have this epic be a summer worth of
> mini-projects for a single GSoC student, isthat correct?
>
>
>
> On Fri, Jan 26, 2018 at 7:46 PM, Matthias Boehm <mboehm7@gmail.com> wrote:
>
> > just FYI: I've created https://issues.apache.org/
> jira/browse/SYSTEMML-2083
> > with the gsoc2018 label. If you have additional project ideas, please
> file
> > the respective JIRAs. Thanks.
> >
> > Regards,
> > Matthias
> >
> > On Mon, Jan 22, 2018 at 12:13 AM, Matthias Boehm <mboehm7@gmail.com>
> > wrote:
> >
> > > yes, that is a good idea and we should leverage this opportunity. I'm
> > > happy to mentor a project as well, specifically on parameter server
> > > architectures for distributed deep learning in SystemML.
> > >
> > > Right now we can emulate synchronous parameter servers with parfor, but
> > > there are other architectures like asynchronous, stale-synchronous,
> > > heterogeneity-aware, and decentralized, which we should support as well
> > in
> > > form of a dedicated runtime infrastructure similar to local/remote
> > parfor.
> > > If done right, we should be able to leverage common primitives in
> parfor
> > > and these parameter servers.
> > >
> > > Apart from better support for distributed deep learning, this would
> bring
> > > us much closer to the goal of a unified framework for large-scale
> machine
> > > learning as we would be the only framework that supports data-parallel
> > > (distributed ops), task-parallel (parfor), and model-parallel
> (parameter
> > > server) execution strategies in a single system.
> > >
> > > I'll create an epic with sub tasks later this week. The goal would be
> to
> > > integrate these parameter servers with the underlying data-parallel
> > > framework to have a common runtime for distributed operations and
> > > automatically support all underlying resource schedulers such as YARN,
> > > Mesos, and Kubernetes. Since the project is relatively large, we might
> > need
> > > to split it up into multiple GSoC projects along with multiple mentors.
> > >
> > > Regards,
> > > Matthias
> > >
> > > On 1/22/2018 8:53 AM, Nakul Jindal wrote:
> > >
> > >> Hi Krishna,
> > >>
> > >> That is a great idea. Thank you for offering to be a mentor/co-mentor.
> > >> I suggest that you think of a list of projects that you can mentor,
> > >> discuss
> > >> them on the mailing list and we can add them as JIRAs with the
> > appropriate
> > >> tags (as required by Apache).
> > >>
> > >> If there are students on this mailing list who want to suggest a
> topic,
> > >> that is very welcome too. If a mentor is willing to take it on, they
> > will
> > >> respond either here or on a JIRA.
> > >>
> > >> -Nakul
> > >>
> > >>
> > >>
> > >> On Sun, Jan 21, 2018 at 9:49 PM, Krishna Kalyan <
> > krishnakalyan3@gmail.com
> > >> >
> > >> wrote:
> > >>
> > >> Hello All,
> > >>> I was wondering if system-ml community was planning to participate
in
> > >>> GSoC
> > >>> this year. I had a wonderful time last year and learned a lot!.
> > >>>
> > >>> If we could have a couple of JIRAS for GSoC 2018 it would be great.
I
> > am
> > >>> willing to help out as a mentor/ co-mentor if that is a bottleneck.
> > >>>
> > >>>
> > >>> [1] GSoC Website (https://summerofcode.withgoogle.com/)
> > >>> [2] GSoC Time Line (https://developers.google.
> > >>> com/open-source/gsoc/timeline)
> > >>> [3] GSoC 2017 JIRA (https://issues.apache.org/
> > jira/browse/SYSTEMML-1451)
> > >>>
> > >>> Regards,
> > >>> Krishna
> > >>>
> > >>>
> > >>
> >
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message