Run the ipython notebook iclr2021_stats.ipynb for the results below. Alternatively, play with the authors & affiliations data in their respective csv files. I have also included a direct script for scraping OpenReview at scrape_openreview_from_csv.py, using simple selenium & beautiful soup methods.
The file iclr_results_final.csv is from Sergey Ivanov's spreadsheet at https://docs.google.com/spreadsheets/d/1n58O0lgGI5kI0QQY9f4BDDpNB4oFjb5D51yMr9fHAK4/edit#gid=1546418007.
Also, take a gander at last year's repo on ICLR 2021 stats by Shao-Hua Sun at https://github.com/shaohua0116/ICLR2020-OpenReviewData. It has a lot of cool viz not included here, particularly related to word clouds. A diff between the two years would be very cool — especially if it's enough to see any shifts. Any guesses? At any rate, a decade is more realistic to see semblance of meaningful shifts.