Finding power laws has now become de rigueur when analyzing popularity distributions. Long tails have been reported for the frequency of word usage in many languages [2], the number of citations of scientific papers [3], the number of visits (hits) to individual websites in a given time interval [4], and many more.
We discussed a few times about whether it is possible to determine the quality of a paper by extracting visual features from the paper and then learning a mapping to a measure of quality such as the number of citations etc. This paper circulated at CVPR 2010, and does exactly that, mapping visual features to estimate whether it has been accepted for the main conference or the workshops.
Amazing how some guys from some other university also did pretty much the same thing (although they didn't use the bidirectional stuff) and published it just last month. Just goes to show you can dump pretty much anything into an RNN and train it for long enough and it'll produce magic.
http://arxiv.org/pdf/1410.1090v1.pdf
LSTMs: that was also the first thing in the paper that caught my attention! :)
I hadn't seen them in the wild in years... My oversight most likely. The paper seems to be getting ~100 citations a year. Someone's using them.