Tuesday, June 17, 2014

Post-PLDI Blog

I had a great time at the meeting this year.  The poster session was a great way to get to meet people, and there were quite a few interesting talks.  I was particularly fascinated by Krishna Palem's keynote talk on inexact computing, and the rest of the talks on approximate computing.  It is a new thought to me, and seems to have some really neat implications for all sorts of problems (e.g. scientific computing, and time-critical decision making problems like weather forecasting or stock trading).  While the notion that we can't rule out certain inputs yielding catastrophically incorrect results (as demonstrated in Eric Schkufza's talk "Stochastic Optimization of Floating-Point Programs using Tunable Precision") is somewhat unsettling, I think that may only be because we are so used to treating computers as fully deterministic machines.  However, we certainly make quite a few decisions based on computations we do not fully understand (particularly in the case of human intuition)!

I also really enjoyed the "Future of PLDI" discussion.  I resonated with a few of the things said, including the idea of making some optimizations on the reviewing process.  In my opinion, having each reviewer handle 20 papers is too much; it cannot allow enough time for reviewers to obtain an in-depth understanding of each paper.  Reducing the number of papers assigned to each reviewer and reducing redundancy in reviewing resubmitted papers could go hand-in hand.  One idea was to allow authors to submit with a list of conferences they would be willing to publish to.  Since authors are likely to re-submit the same work to lower-tier conferences, the reuse of reviewers would save reviewer-time.  Another idea I liked was to pre-publish on a site like arXiv and encourage peer feedback prior to publication.  This would improve the quality of submissions, likely streamlining the review process.

All in all, attending PLDI gave me a chance to meet fellow researchers, and get better acquainted with the community and research at large.  Thanks for the opportunity!

Jake Brock

No comments:

Post a Comment