NIPS submission times

Each paper submitted to NIPS is given a submission number. Here these are plotted against submission time for the points I know for 2003–2008. Note that all times are for first submissions as resubmissions do not create a new submission number. The paper IDs do not necessarily start at 1, or the same number each year.

NIPS started using a different conference management system in 2009, since then times for first submissions are not readily available.

[Graph of submission number vs minutes left until deadline]

Some of the plot is clearer after a non-standard, but smooth non-linear transformation of the time axis:

[Graph of submission number vs minutes left until deadline]

The plot with time on a genuine log-scale is also available. For that, negative submission times were simply discarded.

As suggested by Phil Cowans some time ago, there is also a version with a scaled submission number axis. I scaled the numbers by the largest number observed just before the deadline. This makes less sense now: 1) we know that the submission numbers are subject to an arbitrary offset; 2) in 2008 there were problems with the submission server preventing submissions in the last 5–10 minutes. As a result, the scaled curve doesn’t closely follow the 2007 curve, even though they looked much the same in the original data.

If anyone wants to do anything better with these data they are available: 2003, 2004, 2005, 2006, 2007, 2008. The columns are submission number, date in June, hour, minute, seconds. Times for 2003–2006 are in UTC-0700. From 2007 the times are in UTC as the deadline shifted to midnight UTC from midnight UTC-0700.

For some data and analysis for other conferences, see Patrick Flandrin’s publications. Item [C157] is:

P. Flandrin, An empirical model for electronic submissions to conferences, ECCS 2009, Warwick (UK).

Also [C155] is a French version.