Dear colleagues,
We are happy to announce second thematic session on reproducible
research
in computer engineering that will be held at the spring HiPEAC
computing week
in Paris on the 2nd and 3rd of May, co-located with ACM ECRC
2013.
Based on your feedback and suggestions from the first thematic session,
we are starting building an academic and industrial workgroup interested
in:
1) setting up practical public repository and infrastructure to share
and reproduce experimental results from the community
2) sharing common benchmarks, codelets, data sets, tools, interfaces,
predictive models, etc
3) preparing common experimental methodology including statistical
evaluation
and predictive modeling
4) pushing forward new publication model where experimental
results are
validated by the community before being published
(similar to conferences and
journals in statistics, machine learning,
biology, etc) - the main challenge
is how to enable validation or experiments
across multiple ever-changing
architectures, tools, benchmarks and data sets
If you would like to give a short 10-15 min. talk to share your
ideas,
practical experience or need for such collaborative approaches
(both
in industry and academia), please send me a title and a few sentences
on possible discussions (
grigori...@inria.fr).
Since
the space is very limited, please get in touch as soon as possible
to be sure
that we can reserve a time slot for your talk.
As the main outcome of this session, we would like to prepare an
online
document summarizing all ideas and possible actions (roadmap)
to make
computer engineering reproducible. We hope to restore
attractiveness of
computer engineering for new students and engineers
making it less ad-hoc and
more scientific.
We will have 2 sub-sessions during HiPEAC event:
1) Thursday 2nd of May (14:00 - 15:30) - main workgroup meeting,
talks
and discussions.
2) Friday 3rd of May (11:00 - 12:30) - tutorial on the 3rd generation
of cTuning public repository and plugin-based infrastructure
(Collective
Mind) to crowdsource characterization and tuning of computer
systems. We will
release codelets and benchmarks (CPU, OpenMP, CUDA, OpenCL),
data sets,
plugins for off-line and on-line auto-tuning, run-time adaptation
and
modeling from our previous research, application for Android mobiles to
demonstrate online collection and analysis of performance data from multiple
users,
and a new interactive interface for LLVM and GCC.
Registration and participation is free for everyone interested in this
topic,
including non-HiPEAC members!
Looking forward to your participation,
Grigori