At ICRA 2022, Benjamin Rosman delivered a keynote presentation on an organization he co-founded called “Deep learning Indaba”.
Deep Learning Indaba is based in South Africa and their mission is to strengthen Artificial Intelligence and Machine Learning communities across Africa. They host yearly meetups in varying countries on the continent, as well as promote grass roots communities in each of the countries to run their own local events.
An indaba is a Zulu word for a gathering or meeting. Such meetings are held throughout southern Africa, and serve several functions: to listen and share news of members of the community, to discuss common interests and issues facing the community, and to give advice and coach others.
Benjamin Rosman is an Associate Professor in the School of Computer Science and Applied Mathematics at the University of the Witwatersrand, South Africa, where he runs the Robotics, Autonomous Intelligence, and Learning (RAIL) Laboratory and is the Director of the National E-Science Postgraduate Teaching and Training Platform (NEPTTP).
He is a founder and organizer of the Deep Learning Indaba machine learning summer school, with a focus on strengthening African machine learning. He was a 2017 recipient of a Google Faculty Research Award in machine learning, and a 2021 recipient of a Google Africa Research Award. In 2020, he was made a Senior Member of the IEEE.
hey everybody. Welcome to Robohub. So this year, I got a chance to go to
[Edit: ICRA] and while I was there, I was listening to a presentation by Benjamin Rossman. Benjamin Rossman is also referred to as Benji. He’s the co-founder of an initiative called deep learning Indaba and deep learning Indaba is a company that’s based in South Africa.
And every year they host, events across the continent of Africa. They pick a different country. They helped to sponsor all the experts in machine learning, deep learning, artificial intelligence from across the continent to travel to that event and form some of these social community bonds. Um, I was really moved by this.
So here is a spotlight on Benjamin Rosman and the deep learning in daba.
I was actually born in Zimbabwe. My family is from, yeah, my family’s from Ethiopia.
And so when I saw your keynote presentation on doing Deep Learning Indaba. Connecting communities in Africa, that don’t have this type of society that’s built for them. I really wanted to actually reach out to you and just have you explain what this is.
Benjamin Rosman: Right. So maybe give a bit of the history.
We, a group of us who are African and have mainly spent time studying overseas and many people working overseas. Um, you know, we, we know each other at that point about 2017, it’s quite a small community. Mm-hmm and we’re discussing that, you know, there’s really not a lot of representation from the continent at these conferences.
We particularly looked at NeurIPS being the flagship machine learning conference. Yeah. And we were not aware of any representation that had ever come from an African institution. And obviously all proudly waving the flags and representing the continent. We thought this was not a great state of affairs and reflecting on it.
I think, you know, there’s a lot of rarely smart people across the continent. And firstly, we don’t know who they all are. We know. Our Alma maters and the universities we work with, but the fact that the community doesn’t even know each other is the first problem. And then, you know, that’s, there’s no critical mass anywhere that, you know, it might be someone interested in dabbling over here or there, or working with Jupyter notebooks they find online or something.
And so we wanted to see who there was, bring people together and we thought let’s have a, a week long summer school, maybe we’ll find or a workshop at that point. Maybe we’ll find a couple dozen people we’re aiming 30 to 40 people. Who’d be interested in attending. And you know, a few of us will give some talks and we opened this up for people to apply and ended up with something like 750 applicants.
Yeah. From all over the continent. And we thought, okay, wow. There’s, there’s something here. Yeah.
And there’s a appetite.
Abate: How did you even like reach 750? That means you must have reached way more than that.
Benjamin Rosman: At that point, it was like a word of mouth thing.
So, you know, we, we all had our own connections and we’d reach out and say, Hey, we’re doing this thing. Can you pass this on? And slowly try and propagate through the network. And then we reflected on it afterwards and said, okay, we don’t seem to have any reach here or there, maybe we can ask for connections.
Yeah. And then subsequently we put together a mailing list, which was MLDs Africa. Um, and that’s given us now a way to broadcast out when things are happening. And obviously since the events happened, we’ve now got all the social media presence and now it’s becoming easier. But you know, at the beginning, actually in our first report, after the event, we had like the last few pages, we just had a list of the people we knew kind of categorized by institutions and countries and so on.
And it, you know, we could fit that on a few pages and now it’s. Now I like it’s, it’s a huge community. Yeah.
Abate: And so for all the people who applied to it, what has been the thing that drew them to, to go cause they’re, they’re having to fly across multiple countries. Um, and you know, depending on what country they’re in, it’s not necessarily easy to travel to… I think your first one was in South Africa.
Benjamin Rosman: Yeah. The first two in South Africa. Next one was in Kenya. And now we’re going to Tunisia this year, which I guess is the other extreme and yeah, I mean, there, there’s a few things, one that for many people they’d never had the opportunity to be at an event like this.
Well, there hadn’t been events like this. And even in say South Africa, where I think we’ve got the biggest community of people working in kind of tech space? Well, when I say tech space, I mean mainly in research. Um, and again, our focus was largely in machine learning, although trying to reach out, but that seemed to be the common point that would attract most people.
For, for people, this is really expensive. So we spent a lot of effort in trying to raise funding in the bulk of our funding to actually fly people over. Um, cuz that is a huge thing. But for many people like they’d be working in [00:05:00] isolation. So a lot of our applications would have these stories that people have written up about what they do.
Benjamin Rosman: And you know, let’s say I’m interested in reinforcement learning and I’ve been, you know, I’ve watched all these videos and read these papers and implemented this stuff and we’d look at it and go like, "wow, some of us haven’t even done that". Interesting. But this person says, you know, there’s nothing like this at my university.
I want to do projects in this and no one can supervise me and yeah. And I just feel so isolated and we get this from every corner of the continent. Mm-hmm and so like, that’s a big draw. Um, yeah, but I say even people that were at well established universities that had machine learning groups. They tended to be small and like disconnected from the community.
And so I think that was a huge incentive for people, but we always had these stories of people writing to us. We obviously couldn’t fund everybody. Um, I think the first year we funded about 70 students completely it’s free for students that are accepted to attend, but like the funding would include flights and accommodation and so on.
Um, but we would still have people that would email us from a place like Ethiopia and say, yeah, you know, they’re so excited to have been accepted, but like a ticket to South Africa costs three times my father’s salary and these kind of things were completely heartbreaking. You know, there, there was this passion.
So we realized early on that, what was important is the knowledge dissemination. So we’ve got this culture of any of the lectures with film we’d put them on YouTube. Um, a lot of the tutorial materials we’d we’d make available online as well. And we’d encourage people to go back and spread stuff to the communities.
In fact, now with our application processes, we’ve got questions around, "how will you further disseminate this to your local community?" Yeah. And like, that’s become a core principle cuz we can really only take a certain amount of people at a week long summer school. Yeah. Um, you know, capacity is limited, but yeah, we want this to reach as many people as possible.
Abate: Ultimately it’s a one week program and then the real value that you bring is what you do the other 51 weeks of the year. So it’s like -exactly- going back and then they’re taking these connections and now they just have something that’s a little bit stronger.
Benjamin Rosman: Exactly. And actually one of the core components, cause we launched a whole lot of other programs around this.
We, we try and experiment a lot with the way we do things. Yeah. And I think one of the most successful programs we’ve had is what we call Indaba X. And this is… The idea was we had a little bit of money left over and we scraped together some of it, and we wanted to support events in other countries. Now it wasn’t sustainable for us to organize these things.
And we don’t know what the best thing for like the community in Namibia might be verse in Zimbabwe. So we had this call for people to apply, to host an event, and we didn’t really specify what needs to happen. Um, and we had a whole lot of applications where there were multiple from one country. This was a great opportunity to bring people together and say like, Hey, did you know these other guys exist in your country?
Yeah. But they would propose their own events and we’d give them some financial and a bit of marketing support. And these ranged from like one day events to week long events. There were some that were, that subsequently been virtual and spaced out and like very interesting ways. People look at this and they range from about 30 attendees up to 300 attendees, but driven by that local community.
And this is a really important way to extend that reach. So that’s now been 30 plus countries that have hosted these events. Ranging from Somalia and Sudan down to South Africa. Yeah. And that’s exciting to see that this is happening everywhere.
Abate: Country to country. It’s not even the same. Cultures are super different languages are very different.
Benjamin Rosman: Exactly. and there there’s a few languages that are overarching, you know, if we, we try and target like English, French but then Arabic becomes a big thing and, and, you know, and then breaking down into, to other local groups.
But you, you find like the, the interests are different in different countries. So for example, we got a big research focus in South Africa. Obviously there’s other things that go on as well, but that’s where there’s a bit of a bias.
Whereas you look at say Kenya, for example, they get a massive startup culture. And so the kinds of things they would want to discuss in the event are gonna be different. And if you’re coming from the outside, you, you don’t know what this is, which was kind of the point of the Indaba in the first place that, you know, coming from say, Europe or the us, you don’t really know what’s happening on the ground and you want, you really wanted to relate to the people there, and this ranges from your theoretical research to more applied research.
And now like also with countries being at different stages of their development and adoption at a government level, in some places, policy discussions make a lot more sense than others.
Abate: Yeah. I mean, speaking from personal experience. In Ethiopia there, isn’t a very large amount of even software developers. The, the amount of jobs for it is also in correspondence with that. and then this is something that must change very drastically country to country.
So what’s the way that people learn about deep learning when you don’t have some of these more fundamental and core building blocks [00:10:00] as parts of the communities locally, already.
Benjamin Rosman: So this is changing. We, we are getting more happening at the universities, but actually one of the gaps we’ve noticed in our applications, one of the weak points is usually academic faculty. And that’s something that we need to change.
And I think a lot of these younger students that are really passionate about it, hopefully will start going into some of those roles. Um, also a lot of people that now have maybe their undergraduate or masters through the networks we’ve established are now able to go and maybe do a PhD abroad and many are interested in coming back or, or go an intern somewhere and then come back.
And I think there really is a strong drive to do this. And I think this will start changing things. Over time, but in terms of how people are getting access to the content, one is online. There’s just insane amounts of really good material online. And that’s, that’s been critical and then there’s other initiatives.
So there’s a, a group Data Science Nigeria that we work with. Quite a lot as the Indaba and their idea is to build up the AI community in Nigeria. And part of what they’d done was basically distributing flash disks with like just a ton of material on everything from some of these online lectures to materials and getting going with tensorFlow or pytorch and like distributing this to their local community.
Even when there’s maybe limited access to the internet. So people coming up with some quite innovative solutions to the problem.
Abate: Yeah. Is that something that, you guys are also thinking of doing, just being able to give these, give these different communities, like, Hey, here’s a list of 10 resources that you should keep your eye out on because even if your university is not necessarily keeping you up to speed, you can keep yourself up to date with the latest.
Benjamin Rosman: So I think the big thing that we are pushing is the community effect that like you should know people that you can talk to.
And, you know, a few examples of this one. I’ve had a few students that from Sudan and in their fourth year of study, they had the, their research project and they got a hold of me and we’d met at the Indaba or something and said you know, we don’t have, we got great local support and supervisors, but people that don’t really have this experience.
Um, and so, you know, I got involved in co-supervising some projects there and some of them were even around doing deep RL on stimulated robotic arms. And there were all sorts of issues with compute and there there’s even sanctions on their country. So it makes a lot of these things difficult, but some of them have now written papers from their fourth year project, which is just super cool.
So having that connection means they can now reach outside of their country to get some guidance, which I don’t think there were any avenues for people to do this before. And then there’s other sorts of initiatives and. The one I like to talk about is Masakhane. So this started from some people meeting up at the Indaba that were interested in natural language processing and particularly on African languages.
There’s lots of interesting aspects of this. There’s about 2000 languages in Africa and many are under resourced. So you can’t just train your models you know, fairly naively and translating from one to the other. Yeah. Um, and this started up as a kind of loose collective and they started having weekly meetings and now they’ve got this really active discord and they’ve been training our papers at a lot of big conferences because there’s like these unique aspects to their problems.
And so there, if anyone’s. Interested in NLP in Africa and they find out about this, they just join this collective. And so there’s these networks that are forming on more specialist topics now, similar things helping happening in the healthcare space. Um, and that gives people the access to whatever technologies they want as well.
so besides natural language processing, are there any other things that you saw, like a unique take on that um, that that’s very unique to the African communities.
Benjamin Rosman: So a couple things come to mind. One is that obviously there’s a, a big agricultural focus mm-hmm
And so we’ve seen a lot of Projects. So during the Indaba, we typically have a poster day, an African research day where we encourage all the attendees, particularly the students to bring posters and present them and have discussions. And actually there’s usually some phenomenal work that’s presented there.
Um, but a lot of the focus is around agricultural applications. Um, and, and something I thought was really exciting was things like a smartphone app that you can take a picture of a casava plant that’s got some disease and diagnose it and, you know, useful tools like this for farmers.
Abate: And these are like unique plants to those farmlands.
Benjamin Rosman: Yeah, exactly. And so this kind of thing of like tools to help local farmers that are maybe even just subsistence farmers you know, inject some intelligence and the effectively your machine crowdsourcing to help them do what they’re doing better.
So there’s these kinds of applications. There’s a lot of applications around things like malaria tracking where, what areas are becoming more or less prevalent. Um, so there’s some things on malaria which connects into [00:15:00] healthcare in general. Um, we’ve got these big issues across the continent where you’ve got a lot of like, really good expertise, but it’s very centralized in cities.
And so if you’re in a rural community, there there’s all sorts of challenges. So a person might go to a local clinic to get diagnosed for something. Some tests get taken. This takes a few days to work its way back to a big pathology lab in, you know, the nearest main city, which may be hundreds of kilometers away.
And then there could have been something that was an urgent problem that needed to be dealt with. But in some cases it’s too late. And so, you know, applying different kinds of machine intelligence to kind of shorten that timeframe.And there’s other sorts of problems like logistics challenges. You’ve got a lot of little farmers in an area. Can you develop techniques for them to bring their produce together, to get to lower logistics costs and maybe have a, a stronger bargaining force in getting good pricing for your products?
Yeah, these kinds of challenges, which I, I, many of them are sure exist in other places, but it’s not a, a priority that people look at.
Abate: There’s a series of different environment settings that, that change, like whether or not a product is viable. So like the cost of labor is much lower so that has a big effect On like whether or not you want to like automate certain things actually that’s a big thing, right? It’s it’s not a very compelling story to say we’re automating this process so that humans don’t have to do it. And when you’re sitting with some countries with like 30 plus percent unemployment, like that’s not going down well, but there’s many jobs that are very dangerous there’s conditions that are causing loss of life and so on.
Benjamin Rosman: And these are problems we need to look at. Yeah. But at the same time, I feel it’s important not to just say, we’ve got these technologies developed in the west and we’re gonna use them to exactly solve African problems by the same token. Like we’re doing some of the fundamental research as well and being an active part in the community.
Cause I think it’s important to have that like two way exchange of ideas. Yeah. And that’s starting to grow up as well.
Abate: Yeah. What you see a lot, is that the, a lot of the technologies that are made say in the US are not necessarily adopted across the continent as a whole, because they don’t really fit those user requirements.
Like exactly phones is a big example. Most of the phones made there are either locally made or made with firms that are specifically making it for those societies.
Benjamin Rosman: Yeah. What’s super interesting is that there are a lot of smartphones. There’s a huge smartphone penetration across Africa, which means you get to think about these kinds of apps, but other kinds of infrastructure aren’t there.
And so, as, as you said, like you can’t just take technologies developed elsewhere and apply them directly. There’s differences in everything from language to what the, the roads look like. autonomous vehicles, for example, The roads are very different.
You plug in some standard off the shelf, like object recognition and firstly, they struggle, the lighting conditions are different. And then the, the kind of construction
Abate: Driving culture.
Benjamin Rosman: Exactly like completely different. Yeah. And, and even like you look at a structure and we had like in, in. Some parts of the one city we’re doing this in the buildings are built in a very particular brick way.
Mm-hmm and the software is just identifying every building as a prison. right. So there there’s like these kinds of challenges that require some local adaptation and thinking about it.
Are there big players in industry that are very impactful and do you partner with them through Indaba?
Benjamin Rosman: So.
I, I think it’s in quite an early stage and there’s a few different kinds of things from the Indaba’s perspective. We originally would say go to all the big tech companies for support, and they’ve been amazingly supportive actually. Um, but we didn’t wanna give this impression that, oh, just if you want to be in this field, all the exciting things are off continent.
Um, there’s more and more happening with the startups scene, so like the early ones, there were machine learning consultancies and, and that sort of thing. And there’s, there’s quite a lot of that happening then a lot of the banks and the big corporates on the continent.
Benjamin Rosman: Some of the telecoms companies are being quite innovative and that’s where a lot of people are working and they’re starting to support the community more and more cause they realize that they need to up the game if they want to attract talent. But you know, there, there’s not much in the way of big companies.
There’s some stuff happening in the drone space because that’s, you know, huge areas of land that you might need to monitor or keep track of crops or cattle or something like that. But really these things are just starting to develop, you know, some of these smartphone type technologies might spin out into small companies, but nothing that’s a, a huge player in the space.
Abate: Awesome. Thank you.
Benjamin Rosman: Great. Thanks so much.