InstaDeep is very pleased to announce that it has had a paper accepted at this year’s NeurIPS conference. The paper’s lead authors are Rihab Gorsane, Omayma Mahjoub (former intern and now permanent member of the research team) and Ruan de Kock. Rounding out the list of key contributors are Roland Dubb (an intern at the time of writing the paper, now completing his Masters at the University of Cape Town), plus Siddarth Singh and Arnu Pretorius of InstaDeep – a clear showcase of how the company actively supports interoffice and academic collaborations as part of a global research department.
The acceptance not only signals the outstanding quality of the team’s research, but also marks Rihab as both the first-ever African female lead author, and first-ever Tunisian lead author, at the prestigious machine learning and computational neuroscience event.
“Towards a Standardised Performance Evaluation Protocol for Cooperative MARL” comprehensively reviews existing Multi-Agent Reinforcement Learning (MARL) research with a focus on the cooperative MARL setting. The authors identified certain trends and concerning inconsistencies that are affecting the current state of research; and proposed recommendations and a standardised performance evaluation protocol that will aid future research and progress in the field. They plan to develop the research over time, by repeating the experiment with more people next year, showing the confidence in the work and team (which will be enlarged and developed).
The work will be presented at the 36th NeurIPS conference this November, and the team have made several resources available immediately:
- The paper, available at https://arxiv.org/abs/2209.10485.
- A website which contains a summary of the research, plus links to the dataset and analysis notebook.
- Given the collaborative and community-focused nature of the project, there is also a form to submit any questions, suggestions, ideas or improvements to help raise the standard of evaluation in MARL.
- A Github repo on a collection of review papers for MARL and evaluation methods in general https://github.com/instadeepai/awesome-marl
- An open-source library containing the code of the main work
Co-Founder and CEO Karim Beguir, reacted to the news, “Everyone at InstaDeep is delighted for the team who worked on this paper. This is a wonderful validation of our vision to empower African researchers in Africa and I heartily congratulate all involved. I also commend Rihab on her ground-breaking achievement – she is a real inspiration for African women researchers in particular.”
Rihab said of the announcement, “This amazing outcome shows what the African continent can achieve when we work and learn from each other. I am honoured to be part of this wonderful team and look forward to continuing our internal collaboration, as well as strengthening our external partnerships with institutions like the National School of Computer Sciences of Tunisia (ENSI) and the University of Cape Town in South Africa. On a personal note, I have learned so much about MARL, and I firmly believe this work will lead toward a more unified consensus on MARL evaluation. I am also extremely happy to have my first full conference paper accepted at NeurIPS!”
Rihab is no stranger to excellence in AI. She joined InstaDeep in 2018 after graduating from ENSI with a degree in software engineering. The very next year she became the first female Google Developer Expert in Machine Learning in Africa and the Middle-East; and continues to build on her self-taught skills in her day-to-day work on NP-hard problems using Operations Research approaches and Deep Reinforcement Learning (RL) techniques, such as the highly complex, combinatorial problem of large-scale train routing and scheduling for a national rail operator. A keen advocate of giving back to the AI community, she participates and mentors at AI events such as Deep Learning Indaba and AI Hack, and also spent several years as a mentor on the Deep Reinforcement Learning Nanodegree run by Udacity.
But it isn’t just Rihab making a stand for women researchers – her co-author Omayma was just as pleased to see the work completed and published. And as someone who started as an intern and is now a fully-fledged InstaDeep team member at the start of her career in the MARL domain, Omayma brings a unique point of view to the achievement: “I am thrilled that our research paper has been accepted at NeurIPS. During my internship, I learned so much and could see the importance of research within InstaDeep, but now as a full-time member of staff, I really understand how the company actively supports the department to achieve great things in their chosen specialities, and I am ready for my next challenge!”
Enabling Africa to become an AI powerhouse
This “giving back” mindset mirrors InstaDeep’s aim of helping Africa and Africans actively shape an AI-powered world. The first NeurIPS paper by a Tunisian lead author is another step in demonstrating how Tunisia can be a centre of excellence for AI research, a vision championed by InstaDeep Co-Founders Karim Beguir and Zohra Slim. In the past year alone, multiple InstaDeep initiatives have contributed towards this:
- Making a supercomputing DGX server available for Tunisian entrepreneurs and researchers in the AI and ML ecosystem. Part of a joint “AI Hub” initiative, it gives users free-of-charge access to powerful high-speed compute, enabling them to train complex models and datasets, run detailed experiments, perform large scale analyses, and more, to help them realise the next big AI-powered idea.
- Organising AI Hack, the biggest Hackathon in the MENA region. Held in Tunis in August, 1000+ participants competed over three days. Participants from 24 countries challenged themselves in individual and team competitions and skills-building sessions; working with experienced on-site mentors, AI experts and advisers.
- Co-organising Deep learning Indaba 2022 in Tunis. Deep Learning Indaba enabled participants from across Africa and the Middle East to take workshops, lectures and tutorials all aimed at improving and updating skills, as well as meeting and networking with their peers and prominent industry names and companies.
All these initiatives have been carefully designed to minimise major cost or logistical barriers for participants, in keeping with the company’s vision, and particularly to stimulate the AI and ML community following the enforced break due to the Covid-19 pandemic.
Recognising AI innovation
InstaDeep is also frequently recognised as a leader in AI innovation more widely. It is one of only eight companies to have been ranked among the global top 100 AI startups for the third year running by industry analysts CB Insights, and was also named as one of the “100 most promising B2B companies in Europe 2022” by FT-backed Sifted in 2022.
Both company co-founders have also received multiple recognitions from industrial, technical and diplomatic institutions for their longstanding and consistent contributions to AI, innovation and work in elevating African talent.