An information compressor in a pale blue dot.

An early overview of ICLR2019 (part 2)

Decisions for ICLR2019 are out! So it is time to take a look at this year’s trends! Which researches have been most successful? Which institutions? Which papers?

(Link to part 1)

This year, 24 oral papers have been accepted, 476 posters, and 918 rejects. This is, 500 of 1418 papers have been accepted, and thus reaching an acceptance rate of 35.26%.

This is the updated table with all papers and decisions:

Let’s now see which have been the most prolific institutions.

Top prolific institutions

The top institutions haven’t changed much since the last edition, in fact, most of the top prolific institutions have grown even more in the number of submissions this year. Most urls such as and have been merged, but please do not hesitate to leave a message if you find an error. In the following graph, I have plotted the difference between 2019 and 2018.

Notice that the number of rejections usually grows with the number of accepted works.

Top prolific authors

Here is the list of the top (co)authors.

Differently from last year, Sergey Levine has surpassed Yoshua Bengio, although Yoshua appears in an oral paper. Anyway congratulations to all authors in the graph. I would also like to mention that many first-authors do not appear in this figure, since they spent most of their effort in a single submission.

So, what is the secret sauce? What should I put into my paper to get accepted? Here are the keywords most related to the different final decisions:


So let’s work on Theory about robust graph neural networks for domain adaptation!!!!

The next two sections are best seen on a computer.

Collaboration between institutions

In the last edition, we noticed that the most prolific institutions where those with more collaborations. How about this year?

So the trend continues, google, microsoft, mit, cmu, berkeley, are the ones that collaborate more with others.

Collaboration between top authors

Let’s see how authors collaborate. This time I will use a graph to see if people forms clusters.

Effectively, some clusters can be seen. For instance, the MILA group is represented by the connections between Yoshua, Aaron, and Ioannis. There is also the Berkeley group with Chelsea Finn, Sergey Levine, Pieter Abeel, et al. Deepmind is represented by the region with Oriol Vinyals, Tim Lillicrap, Nando de Freitas… Interestingly, there is an isolated group formed by MSR researchers, with Tienyan Liu, Tao Qin, and Wei Chen.

I hope you have found this blog interesting!

About the data

The data was obtained from openreview, using their library.

Newer >>