In the Spirit of Masakhane! : Post Deep Learning Indaba Nairobi
With recent advancements in artificial intelligence, it has become paramount to ensure Africa participation is visible, in efforts to ensure diversity and inclusion in creating real life AI systems. One effective way of doing this is establishing foundation of an African AI innovation ecosystem, which starts off by providing technical programmes led by industry experts and showcasing how to implement their theoretical knowledge as solutions in code.
Deep Learning Indaba is Africa’s largest annual machine learning conference. Started off in September 2017, Indaba aims at strengthening African machine learning by executing technical programmes led by industry experts, and from this it has become an yearly pilgrimage to converge and share knowledge on ML advances as a community. DLI was first held in Johannesburg in 2017, receiving over 750 applications and being able to accept only 300. This year’s Indaba received an immense 1300 applications, and after outstretching their limits they could only accept 500 to attend the conference. It is quite clear that there exists an even bigger community that wants to take part in this revolution, and Indaba will be working to accommodate even more of this passion and talent into their future events.
Organized by Nairobi AI and in line with the theme for 2018 Indaba: Masakhane! meaning ‘we build together’, Post Deep Learning Indaba Nairobi 2018 was a one-day meetup held at iHub Nairobi to explore the latest that was discussed during Indaba 2018. The event featured technical sessions from attendees, organizers and honorable mentions of the Indaba. As a pre-run to both Indaba X Kenya and Indaba 2019, Nairobi AI aims to share in Deep Learning Indaba’s vision by fostering and building an AI-centric community in Nairobi in efforts to position Africa as a key player in the AI era.
Keynote: Indaba 2018 & the vision of Masakhane!
Chris Barsolai, Lead Organizer for Nairobi AI, led the keynote by highlighting the recent developments around the African AI ecosystem. Community is key, and Barsolai noted how events such as Deep Learning Indaba and Data Science Africa have built up communities of researchers, data scientists, and AI developers who are passionate about shaping the future of the continent in the AI era. According to Barsolai, it all boils down to building AI products for Africa by Africans.
With Indaba 2018 theme as Masakhane, Deep Learning Indaba has set off a wave of innovation and advances in the African AI & ML space. Over the year, we have witnessed a major increase in the participation of African countries in research around data science and machine learning. This is the vision that was intended by community events such as this.
Barsolai put out a challenge to the attendees, pushing them to increase Africa’s visibility in the advances being made. He expressed his confidence in what African participation could spell out. Continent-wide, there is immense talent that could be empowered to serve cross-sector and employ an impactful approach to developing the world around them using deep learning.
Barsolai also recapped on Indaba 2018 highlights, from the poster sessions, technical programmes & codelabs to the continent-wide developer community attending the week-long conference.
The Indaba organizers were well recognized and appreciated, having put in efforts to arrange an event as large as Indaba. Special thanks to Shakir Mohamed and Muthoni Wanyoike for supporting community initiatives all over the continent.
Sessions
Basing off of Indaba 2018’s sessions, Post DLI was structured to capture and convey what we deemed the most important and fundamental topics relayed at the main Indaba conference. Below are highlights of each of the sessions that was undertaken:
CONVOLUTIONAL NEURAL NETWORKS
CNNs have been extensively used in areas such as image recognition and classification over the past decade. Gerald Muriuki delved into the principles of CovNets such as sparse connectivity and transtion invariance, and how a ConvNet is able to recognize scenes.
He explained how local connectivity reduces the number of parameters in the whole system and makes the computation more efficient, and gave examples of how to build simple classifiers.
Slides: https://goo.gl/Vmhii4
GENERATIVE MODELS
Miles Obare, a data scientist at Loanbee steered this session where he delved into generative models. As subtly explained, these models help us to learn the probability distribution from which the training data was sampled from. Miles started by covering PixelCNN/RNN, then looked at Variational Autoencoders and concluded with Generative Adversarial Networks. Throughout the session, the audience was given nice visualizations of how the models work and the expected outcomes. Finally the audience was guided on where they could get more information on the topics covered.
Slides: https://goo.gl/o1yDiT
SUCCESS STORIES OF REINFORCEMENT LEARNING
Reinforcement learning(RL) has been at the center of some of the most publicized milestones of artificial intelligence(AI) in the last few years. In this presentation, Cate Gitau talked about some of the Deep Reinforcement learning systems that have been built by DeepMind, an AI research company based in London and was acquired by Google in 2014. The talk was originally given by David Silver at the Indaba.
Working as a data scientist at Africa’s Talking, Cate explained how fascinated she is on the underlying principles of the ML research area, and the different architectures used today to solve Reinforcement Learning problems such as Q -learning and Policy Gradients.
Among the success stories she highlighted were: TD-Gammon, DQN in Atari, Deep RL in Robotics, Alpha Zero, Dota 2, Capture the Flag and the most famous one, Alpha Go.
Slides: https://goo.gl/LYaypm
GETTING INTO ML RESEARCH & PAPER WRITING
The most interesting session was conducted by Albert Kahira, who joined in via Skype and conducted his presentation all the way from Barcelona. Albert is a PhD student in Computer Architecture at Barcelona Supercomputing Center, and is well vast with HPC memory systems and currently undertaking research on Resilience of Machine Learning Workflows.
Albert touched on several topics ranging from joining Grad School to Writing a research paper. On joining Grad school to pursue Machine Learning research, Albert gave a few tips based on his own journey such as finding a research center/university that aligns with your research objectives and different funding opportunities. On writing research papers, Albert took the audience through the entire paper writing process, emphasizing on the need to make it clear the questions the paper answers and using experiment notes in paper writing. He finalized by encouraging the audience to join Grad school and expressing his willingness to help those planning to join Grad school.
He also gave his key takeaways from Deep Learning Indaba, stressing on collaboration among African researchers, investing in community and rethinking infrastructure.
Community is Key
We are in the first hour of the AI era, and so we should ensure that Africa is well equipped to participate in the advances in research fields related to AI and ML. Toward strengthening African machine learning.
Many thanks to the Nairobi AI team comprised of Ngesa Marvin and irene onyango for their efforts in actualizing this event.
Event Photos: https://www.meetup.com/NairobiAI/photos/29329555/
Follow the Nairobi AI publication and join our community on Meetup to see more updates and stories on the state of AI and ML in Nairobi and Africa.
