We pose a question to you: At the leading machine learning venue, the annual Conference on Neural Information Processing Systems (NIPS), how many accepted papers in 2016 came from research groups on the African continent? Or to be more general, how many accepted papers have at least one of its authors from a research institution in Africa? The answer: zero. And what for the South American continent? Similarly. Zero.
What if we extend this to the last ten years. For African research groups the answer remains zero; eleven for South America. In contrast, at NIPS 2016, the list of accepted papers included more than 150 paper authors from the United Kingdom, while more than 1000 paper authors were affiliated with American institutions. Two entire continents are missing from the contemporary machine learning landscape. This is what we mean when we say that African participation in machine learning is low.
The are many factors that affect these participation levels, including skills shortages, researcher confidence, squeezed funding, infrastructure constraints, and inadequate public policy, amongst others . But the data does show that change is achievable. In 2006, Indian institutions also had zero accepted papers at NIPS, but are now represented by 20 paper authors each year on average. China shows this same trend even more markedly. There is much to learn from the strategies they have used in strengthening their research communities.
The questions we asked in the opening are, of course, highly simplified. We would need to further interrogate the data and any systematic errors it might have. We only looked at accepted papers, not submissions. There are many African researchers who are based at institutions beyond our continent. We have also not looked at other conferences, although for many, long-term statistics are not currently recorded.
Ultimately, our aim is to build a stronger community of African AI scientists, and this metric is simply a way to gauge that strength. Steps towards this change can already be felt. We received close to 750 applications for participation in the Indaba. Our technical programme is filled with speakers who have enthusiastically given their time to our cause. The posters to be presented at our two poster sessions are in the state-of-the-art topics seen at any major conference. And, demonstrating the power of a common cause, the Indaba is supported by universities, government institutions, corporate labs, and startups, all eager to work towards strengthening African machine learning.
There remains a great deal of work ahead of us. But we are hopeful. Over the coming decade, we will tell a very different story.
1. P. Lee. African universities need improved research strategies.
2. Africa Union Commission. Science, Technology and Innovation Strategy for Africa 2024.
We are grateful to the NIPS foundation, and Lee Campbell in particular, for the data we used in this post. Special thanks to Emily Muller who we teamed up with to produce the visualisations used in this post.