Professor Vukosi Marivate of the Department of Computer Science at the University of Pretoria (UP) recently returned from the 2023 International Conference on Learning Representations (ICLR) in Kigali, Rwanda, where he and UP master’s students participated in the conference.

This was the first time that the ICLR, aimed at bringing Africa a few steps closer to the global discourse on the impact of artificial intelligence (AI), was hosted in Africa. It was organised with the help of Prof Marivate who spent the past year collaborating with other programme chairs that included academic and industry researchers in the field of machine learning.

“The International Conference on Learning Representations is one of the top artificial intelligence conferences in the world,” said Prof Marivate, ABSA Chair of Data Science at UP. 

“It is one of the most prestigious conferences in the field, both as a publication venue for the science of AI and because of the high quality of its scientific programme. Given the journey of the African machine learning or AI community over the past decade, having one of the top conferences hosted in Africa for the first time is a milestone.”

UP’s Prof Vukosi Marivate with master’s students Rozina Myoya, Kathleen Siminyu and Thapelo Sindane at the 2023 International Conference on Learning Representations (ICLR).

Prof Marivate delivered a keynote speech at the ICLR’s IndabaX Rwanda workshop, while master’s students from UP’s Data Science for Social Impact (DSFSI) research group demonstrated their readiness to be part of the global move to embrace innovative technology in the quest to solve global challenges. 

In his talk, Prof Marivate highlighted some of the research group’s technical achievements in AI and African language research. 

“I talked about a recent project – titled ‘Preparing the Vuk'uzenzele and ZA-gov-multilingual South African multilingual corpora’ – which uses South African government speeches and newspapers to create new datasets in all 11 South African languages,” he recalled. “With this, we were able to create new translation models and benchmark against other approaches. I used that to dive into the main topic of the policy implications of our research, the data collection process, privacy, security and my recent involvement on the African Union team that is drafting the AI policy for the continent.” 

DSFSI master’s student Kathleen Siminyu delivered two talks, one titled ‘Reflections on dataset building: ideas for benefit sharing’ at the IndabaX Rwanda workshop, and another titled ‘Honouring Kiswahili with technology and community’ at the Practical Machine Learning for Developing Countries workshop. She also took part in a panel titled ‘AI moving forward, leaving no one behind’ at the Women in Machine Learning social.

Rozina Myoya, another DSFSI master’s student, participated in the AfricaNLP workshop by showcasing the research carried out by the DSFSI during its summer recess internship. The study, titled ‘Fine-tuning multilingual pre-trained African language models’, explored whether pre-trained language models can be modified to perform similarly well on different African language data. The goal is to increase the ability to effectively perform downstream natural language processing (NLP) tasks such as text classification, name entity recognition and sentiment analysis in African languages.

UP master's student Kathleen Siminyu giving a talk on her work as a Mozilla fellow.

Master’s student Thapelo Sindane was part of the volunteer team that assisted speakers, coordinated track sessions, set up posters and kept track of the ICLR website. In his free time, Sindane attended workshops, talks and poster sessions to learn about recent state-of-the-art AI technologies.

The conference was meant to be hosted in Ethiopia in 2020, but due to the COVID-19 pandemic, it was held online. By having it in Africa, the participation of Africans at the ICLR increased sharply from 16 in 2019 to 291 in 2023. Several participants highlighted how they were able to attend such a conference because it was held in a country that had a better visa regime than many Western countries that have previously hosted the conference and for which participants were required to get visas.

Prof Marivate said that there were almost 5 000 paper submissions (1 574 accepted papers) and over 6 000 reviewing volunteers (including reviewers, area chairs and senior area chairs). There were tight deadlines and a rigorous double-blind process, he added.

He encourages Africans to actively participate in discussions around AI, its development and its research agenda so that they can play a role in shaping technologies.

Find out more about the Data Science for Social Impact research group