Deep Learning Indaba𝕏 South Africa 2019 — looking back, moving forward
We were thrilled with those who joined us at the Deep Learning Indabađť•Ź South Africa 2019. Judging by the feedback, we have become the premier opportunity for students, researchers, and industry to come together from around the country to learn, share, and grow machine learning in South Africa plus beyond.
- Upload your photos to our album and share those beautiful moments!
- Read the pdf report and see where we went right, wrong, and where we are going!
- Watch the talks you missed, or revisit what you loved, on Youtube.
- We even made a documentary!
- If you are interested in being an organiser for 2020, please join the Discord group as a way to show interest (no commitment by joining).
Thanks to our sponsors for help making this happen.
And thank you to everyone who attended, engaged, laughed, and strengthened African machine learning that much more; the Indabađť•Ź is defined by you.
We encourage you to stay in contact with those you met, share your stories on social media (and tag us @DeepIndabaX_ZA), keep up-to-date on advances, make your own breakthroughs (small or big), and see how we can work together to make the future even brighter.
Table of Contents
- Introduction
- Proceedings
- * Day 0
- * Day 1
- * Day 2
- * Poster Prizes
- * Hackathon — Day [2,3]
- * Unconference
- * Hackathon Prizes
- Feedback and Improvements
- * Opening Party
- * Hackathon
- * Unconference
- * Talks
- * Behind the scenes
- What is next?
- * 1. Increase technical content and level of the talks
- * 2. Ubuntu proceedings of Artificial Intelligence
- * 3. Indabađť•Ź Summer School
- Organisation
- Budget and sponsors
- Conclusions
Introduction
The world continues to change at a rapid pace, with artificial intelligence and machine learning increasingly playing an integral role in human advancements. These technologies are expected to shift power dynamics and shape the future in unprecedented ways in what is termed the “4th industrial revolution”. Africa, in turn, is primed to leverage an ever-growing human capacity in order to make important contributions in the machine learning community.
Born from the desire to diversify this community defining the future, the pan-African Deep Learning Indaba was founded in 2017 to spread knowledge and strengthen African capacity in machine learning. Seeded from the root conference, a diaspora of 27 Deep Learning Indabađť•Ź events are spread across the African continent in 2019, each a local community-driven meeting disseminating knowledge wider and deeper.
Having been established in South Africa in 2017 and 2018, the Deep Learning Indaba will be held this year in Nairobi, Kenya. This presented the South African Indabađť•Ź with the responsibility to fill the large gap left behind and become the pinnacle education and research-focused artificial intelligence conference in the country.
In building capacity for machine learning in South Africa, we sought to create lasting connections, drive innovative developments, and proactively influence how recent advances shape our society. The Deep Learning Indabađť•Ź South Africa, therefore, presents an opportunity for students, researchers, and industry to come together from around the country to learn, share, and grow machine learning in South Africa and beyond.
With nearly 500 applications and a venue that seated far less, picking the top 210 candidates was tough. We proactively targeted under-represented groups from around the country, both for delegates and speakers. We also weighted applicants by their motivation to gain knowledge and their desire to build connections, which are fundamental factors in growing machine learning capacity in Africa. Utimately, we picked people passionate about machine learning, giving a chance for those looking to enter the space as well as those looking to build expertise.
Over 74 hours*, the 210 delegates laughed, learnt, shared, and planted roots to grow machine learning in Africa. With 73 travel fellowships that fully funded students from around the country, we were able to service learners who may not have had the means to attend, exemplifying inclusion and diversity. This melting pot of participants embraced the opportunity to engage with top minds and future talents.
While we were limited to 210 attendees due to space constraints, we strongly believed this material must be made freely available for sharing with the wider machine learning communities scattered across the country. To achieve this aim, we had all the talks filmed (the speakers kindly agreed to share their slides and voices), a photographer to capture serendipitous moments (featured in these pages), and a stream of the plenary sessions for real-time virtual engagement.
Our goal of inclusion and diversity informed our aims to 1) provide a practical introduction to machine learning for beginners, 2) create an exchange for sharing scientific knowledge amongst researchers, and 3) showcase cutting-edge applications, cognisant of ethical considerations, to industry.
In this report, we look back at the event to share great moments and success stories, and to reflect on where there is room for improvement. Importantly, we built a strong foundation for the continuation of the South Africa Indabađť•Ź so that inspired individuals can take the torch and make it shine brighter each year.
Proceedings
See the full programme in this pdf
Day 0
Sunday the 14th of April saw South Africa’s Deep Learning Indaba𝕏 open for registration. Due to the large number of attendees and the packed schedule, this opportunity dedicated some time for mingling before bussing everyone to the opening event nearby!
The opening event had a DJ, dancing, and drinks plus food, friends, and new faces. Many found the start “awesome”, “[a] great way to interact”, “lively”, but also “a bit loud”. Entering a busy 3 days ahead, we wisely capped the official event at 10pm.
Day 1
Early on Monday, we kicked off with a short welcoming address by Maria Schuld, Amira Abbas, and Prof. Francesco Petruccione who was speaking as the host. This lead to the first plenary session, a panel discussion on “Africa in AI” moderated by Christopher Currin and featuring a range of panellists with incredible diversity in so many meanings of the word (pictured above).
While unconventional, having the panel at the beginning of the proceedings set the tone on a number of aspects for the event. First, it showcased the diverse backgrounds of everyone at the event: from academia and industry, Cape Town and Johannesburg, Vice Chancellor and MSc candidate, and many more. Second, it served to define a narrative as focusing on AI and its complexities from our African perspectives rather than Africa from an AI perspective. Third, with disagreements arising from the discussion, it showed the audience that even top people in the field may not have all the answers and can disagree amicably. Finally, it fostered discussions throughout the conference by making conversations and controversy welcomed from the very start.
Directly following the panel was our first keynote speaker Prof Tshilidzi Marwala, Vice Chancellor of the University of Johannesburg talking on the rationality of artificial intelligence machines.
After a short tea break, the proceedings split into three tracks: Foundations; Research; and Industry, Policy, and Ethics.
- Foundations was aimed at attendees early in their machine learning career or for those looking to learn a new technique such as unsupervised learning or reinforcement learning. These sessions were targeted as practical introductions where delegates could follow along on a computer, yet near the end demonstrate some advanced knowledge and powerful applications to entice further engagement beyond the session.
- Research showcased our homegrown research activities across South Africa, featuring representatives of some major groups active in the machine learning sphere. Because of the variety of topics and attendee backgrounds, these talks typically included ample background knowledge before delving into results.
- Industry, Policy, and Ethics exhibited influential minds from diverse companies. The development, application, and impact of AI was poignantly discussed and engaged with by audience members both established and starting out.
The end of the first day saw the spotlight shift from invited speakers to attendees’ research with a poster session.
Day 2
Tuesday morning broke with a keynote by Andrew Saxe, a post-doctoral research associate in Deep Learning, Neuroscience, and Psychology at the University of Oxford. The in-depth insight he gave into the interplay between deep learning and generalisation was closely followed by the audience to form a strong start to the second day of proceedings.
Again the day split into the three parallel tracks until the final talk and keynote by Bruce Bassett, who is a professor of Mathematics at the University of Cape Town/AIMS/SAAO, Head of AIMS cosmology and machine learning group, and Head of Data Science and Machine Learning at SKA South Africa. His fun and adventurous 4-part talk outlining AI developments and its future was welcomed as a fitting ending to the formal presentations.
Poster Prizes
Immediately afterwards, the poster prizes were awarded: 1st place Samantha received a full travel fellowship to the Deep Learning Indaba in Nairobi, Kenya to present her work on Predicting social unrest events in South Africa to 1000+ participants! 2nd place Elan won a full pass to the AI Expo Africa in September (valued at R4500) courtesy of Cortex Logic and 3rd place Simbarashe won the “Deep Learning” textbook courtesy of the organizing committee member Amira Abbas.
But while the day had ended at 17:30, the hackathon was about to begin!
Hackathon — Day [2,3]
After two days loaded with learning, the participants were eager to get their models dirty with data during the HACKATHON. Working with committed sponsors, we could offer the participants a range of problems for them to tackle over 18 hours.
Open Data Durban provided the Indabađť•Ź hackathon team with the South African Census data from 2011 for a beginner challenge we crafted focused on basic concepts such as data pre-processing, regression, and classification. Isazi Consulting challenged more experienced participants to develop a model that used image segmentation and classification to convert images of numbered sequences into their text equivalents. IBM provided a dataset comprising of DNA sequences and tasked the participants to use deep learning to predict whether the DNA sequence was from a zoonotic or non-zoonotic virus*. Finally, the Aerobotics team released a dataset of orchard imagery with the goal of training a Cycle GAN to segment the pixels in images between a set of classes. Sponsor-provided tutors were around to help too!
With the incentive of some pizza (thanks Data Prophet), snacks (thanks James Allingham), some high performance computers (thanks CHPC#), interesting challenges, and amazing prizes up for grabs, the participants worked late into the night at the inaugural Indabađť•Ź South Africa hackathon.
Unconference
The third day began with hackers returning to the UKZN LAN venue as well as many others eager to engage on interesting topics not covered by the schedule during the UNCONFERENCE. Groups formed and converged around diverse topics such as “causality”, “GANs”, “ML and the scientific method”, and more which sprouted throughout the allotted time. This provided a semi-structured form of knowledge transfer that is less formal than a talk, yet with experts naturally leading discussions.
At lunch time, the unconference and hackathon came to a close and in the final event of the conference, we began the hackathon prize giving.
Hackathon Prizes
For the Open Data Durban challenge, 1st place, Lerato Mnguni, won the “Deep Learning” textbook by Ian Goodfellow, Yoshua Bengio and Aaron Courville courtesy of Data Prophet and 2nd Place, Tarirai Chani, won a R1000 Exclusive Book’s voucher thanks to the UZKN Big Data Flagship.
Aerobotics awarded 1st and 2nd place to two teams: Computer-Vision (Ronald and Omran) and No Name Team (Katlego, Ditebogo, Moyahabo, and Crestinah), who each took home an array of amazing textbooks: The Elements of Statistical Learning, Pattern Recognition with Machine Learning, Machine Learning: A Probabilistic Perspective and Reinforcement Learning: An Introduction.
For the Isazi Consulting challenge, 1st place went to Michael Gant and Guillaume Odendaal who each won a GEFORCE RTX 2060 GPU and 2nd Place, Nangue Tasse Geraud, won R1000 in cash.
For the IBM challenge, Mkhupuli Ncube grabbed 1st place with a cash prize of $250, Aerobotic’s own Scott Glover came 2nd with $150 and finally, Craig James Bester and Christopher Dunderdale, placed 3rd with a prize of $100. Special thanks to the Eck Institute for sponsoring the IBM prize.
Feedback and Improvements
After an exhausting event, it was time to rest, digest, and reflect.
Nearly 75% of attendees rated the Indaba𝕏 as a 9 or 10 (out of 10) overall, with many finding the content “very relevant” and the speakers “knowledgeable” with “some great insights.” Attendees had some powerful key moments from the event that they shared including ,
South Africa has a thriving and diverse machine learning landscape
Tshilidzi’s keynote, the panel as well as the research talks.
networking … really motivated me…
as well as some inspiration that these events do make a difference:
hands down the best experience of my life!
I will forever be grateful for the opportunity to learn and grow
While much of the logistics seemed well-rated, the registration process in particular, some thought the communication could have been better (e.g. sending the schedule earlier), the venue needed more power outlets, and the accommodation and transportation needed greater attention for those who received travel fellowships. While we were fortunate to secure sponsorship to fund 73 students from around the country to attend, with special acknowledgement to the DST-NRF CoE-MaSS* grant, we could have done better.
* Department of Science and Technology (DST) — National Research Foundation (NRF) Centre of Excellence (CoE) in Mathematical and Statistical Sciences (MaSS)
Opening Party
The opening party was well-received with many enjoying the food, atmosphere, and spontaneous conversations. The main criticisms were that it was too loud for some and that with no ice-breakers, some cliques formed.
Hackathon
In 2018, the Indaba𝕏 Western Cape team had investigated organising a hackathon, but it was not to be. One of the key lessons was how long it takes to get data approved (even if it is publicly available). The Indaba𝕏 South Africa team therefore started the process at the end of 2018 (before official Indaba𝕏 announcements) by talking to companies with which to partner. This worked well as 70% thought the hackathon was great (8 or higher). Some ”hackers” did think all the problem were too advanced, yet we did try to include educational aspects in some of the problems. In future, it would be good to better align the hackathon sponsors in terms of prizes and starter-code expectations.
Unconference
The unconference was a novel addition to Deep Learning Indaba𝕏 schedule as it is a relatively new concept of having the audience drive the agenda and structure the content. While some participants found the unconference discussion on causality a highlight, others didn’t find a critical mass in their own topic of interest.
We had asked the attendees to provide suggested topics for the unconference beforehand and then decided final groups of discussion on the day. With a diverse array of topics which came up, it was difficult to narrow down the suggestions.
Advice for future unconferences, which we think should be included if given the time, is to have more time at the beginning to distil topics of interest, create a more structured agenda where people can move between topics at intervals, and have a few more leaders that can guide the event.
Talks
The talks were very well received by the delegates with particularly high scores achieved by the industry track. All tracks were roughly equally popular with the attendees, and the keynotes were met with overwhelmingly positive feedback.
An important point of internal criticism, coming from the organising team, is the diversity of the speakers. Although we put a lot of effort into racial and gender balance, with the result being much ahead of any international machine learning conference, we hope to feature an even more diverse line-up in the future. A way to encourage this would be to allow more MSc/PhD students to present. Furthermore, we have started an initiative to have a shared pool of speakers with the other 26 Indabađť•Źs to help achieve a broader representation.
Behind the scenes
While overall a success, we learnt some important behind-the-scenes lessons along the way including the difficulty in handling finances partially based on university cost centres, organising ground transportation for sponsored students, and ensuring the commitment of sponsors and the host institution. To improve next year’s organisation handling of finances, we have set up a non-profit to facilitate more rapid payments.
What is next?
1. Increase technical content and level of the talks
Balancing the different levels of technical backgrounds is a major challenge for a conference that understands itself both as a platform for excellent academic exchange, but also as a catalyst to form a community. We feel the focus on accessibility of content was well-suited to this type of event, but some clearly want to see a more technical-focused research conference, in the vain of the big international conferences, in the near future. We believe that the South African Indabađť•Ź should aim to steadily increase the technical content and level of the talks; to grow the content with the growth of the academic landscape in South Africa.
2. Ubuntu proceedings of Artificial Intelligence (UAI or “You n I”)
The Indaba𝕏 initiative is a means to experiment with new ways of building capacity in machine learning and so the focus tends to be broad to grow the many groups which make up our community. Given that there are few research-focused machine learning conferences in South Africa, and Africa, it would incredibly powerful to host one harnessing the spirit of an inclusive community — the Ubuntu* proceedings of Artificial Intelligence. One idea would be to reach out to IEEE South Africa with a view for submitted posters, with extended abstracts, to be peer-reviewed and published so that there is further incentive to attend as well as to attract more international attendees.
3. Indabađť•Ź Summer School
The proceedings could stay connected with the Indabađť•Ź education roots with a short and intense summer school beforehand with fewer attendees but higher engagement, akin to the Machine Learning Summer School in Stellenbosch. The attendees can then attend the proceedings for free and gain exposure to a conference setting.
Organisation
A group a highly motivated team players from across the country rose to the occasion of organising one of the biggest Indabađť•Ź events on the continent and one of the biggest artificial intelligence and machine learning conferences in the country. Passion, planning, and patience meant the team worked together to pull off a 3-day event that inspired and motivated attendees with few hiccups.
While spread across the country, and with some having never met IRL*, the modern age has meant remote working is true for conference organisation too. It helped using a central repository of information, having leaders clearly documenting and communicating decisions on a Discord# server, meeting on a weekly video call for group rapport and discussing larger issues, and creating a message group for “firefighting” during the event. Starting the planning early was paramount to the event’s success, especially with regards to organising a hackathon. Local organisers, of the host institution, were necessary to acquire the venue and set the dates during holidays to facilitate sponsored accommodation in nearby residences. Thank you Prof Francesco Petruccione for being willing to host and Dr Maria Schuld for leading the team as a local organiser.
List of Organisers (alphabetical)
- Amira Abbas (UKZN)
- James Allingham (Wolfram Research)
- Annika Brundyn (UCT)
- Alex Conway (NumberBoost)
- Christopher Currin (UCT)
- Steven James (Wits)
- Sasha Naidoo (Nedbank)
- Andrew Paskaramoorthy (Wits)
- Francesco Pretuccione (UKZN)
- Anban Pillay (UKZN)
- Maria Schuld (UKZN)
- Sicelukwanda Zwane (Wits)
Budget and sponsors
The sponsorships were kicked off with the DST-NRF CoE-MaSS, thanks to some Deep Learning Indaba organisers, which meant we knew we could scale this Indabađť•Ź to include travel fellowships, a poster session, a hackathon, and an opening party. While registration fees can offset expenses, it does require additional admin and more uncertainty. Fortunately, we managed to balanced the books almost perfectly (net = -R210.78). See a breakdown in the pdf report.
Without the support of our amazing and generous sponsors, the Deep Learning Indabađť•Ź South Africa 2019 would not have been able to reach its dreams.
Deep Learning Indaba
UMHLANGA ROCKS TIER
- University of Kwazulu-Natal
- DST-NRF Centre of Excellence in Mathematical and Statistical Sciences (CoE-MaSS)
NORTH BEACH TIER
SOUTH BEACH TIER
HACKATHON SPONSORS
- DataProphet
- Isazi Consulting
- Aerobotics
- #CSIR Centre for High Performance Computing (CHPC)
- Eck Institute for Global Health
NEW PIER TIER
- Kwazulu-Natal Research Innovation and Sequencing Platform (KRISP)
- Psybergate
- ConversionScience
- NumberBoost
- Xineoh
COMMUNITY PARTNERS
Conclusions
With the rapid advancement of AI across the world, it is imperative that Africa not only has a voice, but an authority in the field. One way to stand out is by embracing “African problems” through our domain knowledge and unique contexts to solve local problems that impact global solutions. Doing it alone is never an option. The (South) African AI community is growing at an exponential rate year-on-year. The connections formed at these events form a network of people integral to growing the field and each other.
Collaborations between local and international universities and companies are vital to growth and development. In addition, the noticeable number of international African students at the Deep Learning Indabađť•Ź shows hunger for greater partnerships on the continent. Still, inter-country policies need to be addressed to facilitate better communication between collaborators. Fortunately, the Deep Learning Indaba movement is a means to strengthen African machine learning through knowledge transfer and connection building.
These events last beyond the schedule and reach beyond the attendees. The passion of everyone involved creates power. Powerful algorithms. Powerful connections. Powerful people that grow the space and seed the next event.
We truly appreciate everyone involved in creating a special event. From those attending defining the experience to the sponsors supporting the organisers in executing their vision. We are thrilled to see so many people eager to push what is possible and create a brighter future.
From this experience, which we poured our souls into, we have celebrated our successes and learnt from our failures, shared some incredible laughs and formed new or stronger bonds. Now it is time we look forward and see how you can play your own role in making Deep Learning Indabađť•Ź South Africa 2020 that much more special.
