Written by Davit Soselia & Shota Amashukeli —January 31, 2020
NeurIPS is one of the major conferences in machine learning, with interest in it increasing yearly. At NeurIPS20, 12115 abstract and 9467 full paper submissions were received, a 40 % increase over the previous year. It is fairly selective as well, with only 1900 papers and 105 presentations admitted, an acceptance rate of just over 20 %.
This year had a great selection of both theoretical and applied works, with advances in the use of graphs, deeper understanding of language models, practical application of continual learning, etc.
There seems to also be an increase in collaborations, marked by multi-author works. As usual US is main contributor both in academia and industry, with whopping 1186 papers in total. Sergei Ivanov did a in-depth analyzes of paper origins and affiliations.
Interestingly, Alphabet is leading the charge, with Google and DeepMind having the most papers published from the US and UK respectively, while in other countries these spots are held by universities and research institutions.
Still, without accounting for the relative population size of each country it is somewhat tough to get a perspective of what the ML worldwide landscape looks like.
First, let us grab the population and GDP figures for each country. The data we used is available here in a CSV format:
With the number of papers, GDP figures and Population size loaded we can look at which countries lead when adjusted for each category.
First of all, let’s see which countries published the most papers per capita.
Now Switzerland, Singapore, and Israel take the lead with quite a margin, With the USA and UK maintaining leading spots, and the Nordic countries being well-represented as well.
Next, let’s see who gets the most for their buck — Glancing at the number of papers over GDP
Here as well Israel, Singapore, and Switzerland seem to be doing quite well. Cyprus also makes the list with an exciting paper regarding Distributionally Robust Estimation co-authored by Angelos Georghiou, professor at the University of Cyprus.
And finally, when adjusted for GDP per capita we have the following picture:
With China and India advancing to the top spots, publishing most papers considering their relatively low GDP per capita.
Overall, the US is still at the top when measured by most metrics, but other countries are also making advances. With NeurIPS having authors from an increasing number of countries it is our hope that the future of AI will be decentralized and open to all nations despite their wealth.
