Cape Town, 01 Oct 2020 – In 2018, Arnu Pretorius broke ground as the first African researcher to get a paper accepted at NeurIPS (Neural Information Processing Systems Conference), one of the most prestigious AI Conferences in the world.
With “Critical initialisation for deep signal propagation in noisy rectifier neural networks” (arXiv:1811.00293v2), Arnu, still a PhD student at Stellenbosch University, together with Elan van Biljon and supervisors Dr Steve Kroon and Dr Herman Kamper, showed the world that African talent could excel in ML.
And the 34rth Annual edition of NeurIPS cements that fact.
Indeed, Arnu’s second paper “A game-theoretic analysis of networked system control for common-pool resource management using multi-agent reinforcement learning” just got accepted into the peer-reviewed elite conference, making him the only African-based researcher in history to have two first-authored papers accepted at NeurIPS.
“The work is truly a joint effort with an amazing team of students: Scott Cameron, Elan van Biljon, Tom Makkink, Shahil Mawjee and Jeremy du Plessis. Besides InstaDeep, the work is in collaboration with three of the top universities in Africa, namely Stellenbosch, Wits and UCT. I think this really showcases some of the talent we have on the continent and the power of collaboration (between industry and academia). We must keep strengthening this important partnership.” Pretorius said.
The research paper, sponsored and supervised by InstaDeep, analyzes from a game-theoretic perspective how networked multi-agent RL systems behave (under varying constraints) when used for common-pool resource management.
Pretorius, who joined InstaDeep in January of this year as a Research Scientist and team lead in the Cape Town office, is thrilled by the paper’s acceptance and hopes this research may help engineers better understand multi-agent systems, and in doing so, also help them build smarter systems, potentially improving the way common-pool resources are managed in our communities.
“For me personally, it is of course a wonderful achievement as an African-based researcher to have been accepted twice at NeurIPS, and I’m proud to have done this work as part of the InstaDeep team. Not coming from a Reinforcement Learning background, it has been quite a challenge to change my research focus. However, I’m extremely passionate about the potential of RL to build truly intelligent and adaptable systems. Given the current issues we are facing, such as a global pandemic and climate change, these types of systems will become ever more important. Finally, this work would not have been possible without the wider African ML community, sharing their knowledge and helping each other, specifically, the influence of one of our strongest partners, the Deep Learning Indaba.”
InstaDeep was named one of the Top 100 Most Promising AI Startups in the world by CBInsights in 2020 and is widely considered as the leading AI Startup from Africa.