Hi all,

first of all, thanks again for your interest. It's great to see that you
are excited about our GSoC project idea on language and runtime support for
parameter servers in SystemML. I'd like to give a couple of pointers to
clarify potential questions some of you might have.

1) GSoC Guidelines: There are good existing guidelines. Please read them
and make sure that you're eligible.
http://community.apache.org/gsoc.html
https://google.github.io/gsocguides/student/

2) Project Discussion: For all technical discussions around potential
project proposals, let's use the main epic JIRA SYSTEMML-2083. At this
stage right now, it's important to understand the general goal, existing
related work, and how the proposal would differentiate.

Furthermore, think about aspects of the project you would be most
interested in. Examples are (1) language and API extensions for a seamless
integration in SystemML (e.g., paramserv builtin function, Keras2DML
extension, function pointers), and (2) runtime support for
local/distributed, synchronous/asynchronous execution and related
performance features. Both would touch upon SystemML internals but at
different abstraction levels. I encourage you to think about how you would
approach such a sub project. Please don't hesitate to reach out (in public
or private as you prefer) to discuss your ideas.

3) Optional Contributions: If you are unsure about what it means to work on
the internals of SystemML, you might want to consider picking up small
tasks and simply giving it a try. That's perhaps the best way to get
comfortable and to evaluate for yourself if your would enjoy the summer
project. There are many open issues some of which are tracked via JIRAs,
but we're happy to give more pointed suggestions as well.

4) Project Proposals: After working through the technical aspects and your
own ideas, you should be in good shape to write your project proposal. It's
fine to have multiple proposals on the same overall project. At the end of
the day, its your project
proposal that covers your ideas on the how to tackle the overall goal. We
are happy to provide feedback though. I would recommend to start early and
iterate on it.

@Nakul and @Krishna: Please correct me if I missed something (since I'm
volunteering for the first time as a GSoC mentor).

Regards,
Matthias

Reply via email to