Great to see interest in the Indaba from the HN community! This year was a lot of fun; we had some amazing students, speakers, and tutors.
While the practical notebooks we used in 2018 were already amazing, my colleagues and I worked hard on expanding and improving them for the 2019 Indaba. In case anyone wants to check them out, here's a full list of the practicals (the links go to Google Colab notebooks)
We'd love any feedback on the practicals to improve them for next year!
If you have some ML expertise and would like to volunteer to be a tutor at next year's Indaba, which will be held in Tunisia in late August 2020, drop me a mail at contact (at) taliesin.ai. I also encourage anyone to apply to attend as a student!
I don't understand the rather negative introduction in the article. Indaba seems to be about getting people up to speed with the basics of ML, NeurIPS and other ML conferences are about cutting-edge research. Both have an importance place. Why the distinction about Western vs non-Western AI?
This isn't an alternative, it's a way to bootstrap research in Africa. And that's amazing. But the narrative that there's tension between the two and that somehow African AI is "other" and won't find a place in "Western AI" when the research is world-class is totally absurd and does African researchers a massive disservice.
Yes, Africa is under-represented at international AI conferences. But that's a consequence of many things (imperialism, oppression, economics, etc.), not the somewhat-impartial double-blind review system we have.
I wish people would tell a more inclusive and positive story rather than manufacture some conflict where there is none.
Visa policies create exclusion by their very existence. So its not entirely an unfair assumption.
For those who are unaware visa policies are extremely discriminatory to people from African countries. You will not even be granted a tourist visa or even be allowed to attend a tech conference. You are stuck where you are born.
No, I'm saying that people from certain countries have a tough time getting visas because their co-nationals have a habit of overstaying their visa as a form of immigration. Thus making the host country clamp down on issuing visas.
> But the narrative that there's tension between the two and that somehow African AI is "other" and won't find a place in "Western AI" when the research is world-class is totally absurd and does African researchers a massive disservice.
This tension doesn't really come from the specific communities, it comes from the freedom to travel as mentioned in other comments (i.e. it's incredibly difficult not only financially but also in terms of visas and access to countries where conferences take place). That just (unintentionally or not) automatically excludes African "voices", thereby turning them into the "other".
It's therefore a little disingenuous to say that there's no conflict or that it's manufactured; any African academic or tech speaker can tell you there's plenty of conflict in simply applying for a visa to Europe, for example.
>That just (unintentionally or not) automatically excludes African "voices", thereby turning them into the "other".
African researchers can still publish and read in English, it would be extremely unlikely for the two areas of research to evolve independently. Look at American and Soviet research during the cold war - there was a huge language and political barrier, but the Soviets didn't end up with a unique scientific system.
A lot of things did evolve differently in the USSR. For example in linguistics the Soviet Union widely adopted the strange doctrine of Marrism, only to overturn it later.
During the Stalinist era it was well-nigh impossible to cite contemporaneous publications from non-Communist countries, as it was seen as aiding the bourgeois enemy. (It was more acceptable to cite pre-1917 publications from the West, though.) Even after the Khrushchev Thaw, there was still a policy that researchers cite mainly their compatriots, and cite Western publications only when absolutely necessary.
In some fields of linguistics and archaeology, this did result in the USSR taking a very different course in those disciplines than in the West. In fact, the divide is still palpable today, though now it is no longer a matter of not being allow to cite Western works, rather it is down to 1) the influence of elderly Communist-era faculty who by inertia preserve the old restrictions, and 2) Russian universities being unable to afford Western academic books that can cost into the hundreds of euro.
Another interesting example of Soviet influence on the writings of scientists is Lysenkoism, but I would call that a suspension of biological research, as opposed to a change in biological research.
If the results can be changed by changing the underlying philosophy, then what you're doing isn't science. Science will always arrive at the same results no matter how you choose to metaphysically interpret them or embed them in to your national consciousness. Soviet physics and chemistry tracked their western counterparts very closely and later merged back in without any great surprises. Once the central government released its grip on biology it returned to its natural path, the same path that was being independently followed in the west.
I, like the OP, am referring specifically to African researchers' ability to speak in person at conferences in the West. Africans can and do publish research in English all the time, which is sometimes picked up in "conversation" in the wider global community, but without the social emergent properties inherent in face-to-face communication it's difficult to bring that to wider attention (there's a reason tech companies aggregate in SV, for example).
IMHO the beginning of the article wasn't about bootstrapping research in Africa, but was about how researchers there are unable to get visas or afford to travel to AI conferences.
What makes it different from silicon valley? I read the whole article but couldn't find it mentioned except for the headline. Is there something I'm missing? Is it just that they aim to limit sponsorship from silicon valley giants?
That was my question, too! I was wondering if they were developing some qualitatively different form of AI that's not just, e.g., matrix multiplication or something like that. That'd be neat, but I didn't notice anything in the article about that.
It did mention a few African-specific projects: controlling malaria and identifying African wildlife. So, perhaps, it means not that the tech is different, but the applications.
Not adding a source, but Chinese investment on the African continent is huge right now, and has been for years. It has gone largely unnoticed by western media. However, many African nations now owe huge amounts of money to China from large infrastructure projects.
Anecdotally, when I was in Zambia every construction company was Chinese, and every job site was a Chinese foreman with Zambians doing the work.
I didn’t say they weren’t hiring plenty of locals. But they seemed to import the guys running things. The ratio could be 20:1 Local:Chinese or even higher. Basically a Chinese guy saying where to dig, and tons of locals digging. They definitely aren’t bringing in Chinese labourers.
Just FYI that Zambia also has long standing relationships with Russia.[1]
Top government officials in the science space speak Russian.
Also Chinese have been involved in SSA since the 60s e.g. South African Boarder War (aka Namibian War of Independence), which Zambia was also involved in. [2]
Finally most of east and southern Africa was Socialist from 50s/60s until the 90s (many still are), so it makes sense for them to be trading with China, the world power that is closest to their political inclinations. [3]
Other side is UK/US/EU axis, who are the ones who royally fucked it up in the first place (pun intended). [4]
I know who I would choose, if I were in their positon.
When I saw this I was expecting a anti-racist angle. Something like "In SV the white mans bias in machine learning stops minorities from being hired, yet here in Africa... "
This is great for researchers in Africa and the surrounding area, but it seems a simpler solution would be for the CS community to move towards placing greater weight on journals, rather than conferences (like just about all other disciplines). It's not just researchers in the poorer countries of Africa that struggle. Even graduate students in the US can be put in difficult situations where a paper is accepted but they have no travel support from their advisor, department, or institution.
As an anecdote, I recently had to put up $1,500 of my own money to travel to a conference in Italy because my advisor doesn't have any funding and my department only paid for registration. Luckily my wife works and it wasn't an issue paying for the trip, but it was still a frustrating experience.
Let's hope silicon valley doesn't get their hands on them. We need more people fixing real problems for the third world instead of solving uninteresting first world problems.
Good stuff! As part of a tour in Kenya my wife and I spent a day in a village and in addition to people’s kind generosity inviting us into their homes, the other strong impression from that day was how very happy the kids were to be in school.
I took a remote class on globalization from Wharton Business School several years ago and one of the messages was that Africa would become an economic superpower.
While much of the world has benefited from globalization, I like to see each country and region strongly maintain their own culture and national identity. I would find a world that was more or less the same to be a sad thing.
The recent surge of ML research in various parts of Africa is awesome. But from what I've seen (including this article), it is not much of an A.I. industry yet -- most activity is in research with an academic focus. (So, no wonder it looks different from Silicon valley.) Hopefully this helps solve the chicken-and-egg problem needed to get industry going.
While we're at this, there are other domains from which stuff is notoriously being added through a single day and which should already have some restrictions; there's difference from adding interesting content and shameless spamming or as some would prefer "content promotion".
I've joined HN in hope to see interesting IT content, not to be fed by big portals with general news from yet another side.
Medium is encouraging people to put their articles behind a paywall, so many Medium stories are inaccessible unless you log in. Many feel this goes against the spirit of blogging.
I called it years ago: Medium is a magazine with unpaid writers and editors. It was a roach motel to bait everyone into putting their content there with a pretty UI and then paywall it.
Yes but the romantic era of blogging is over. Content creators have to get paid.
I'd rather have paywalls than widespread ads. At least it's the normal arrangement that way: you create something valuable, I pay for it. Very clear cut. With ads that relationship is blurry.
I did not downvote, I was under the wrong impression that all medium posts are under paywall but I checked now ( https://help.medium.com/hc/en-us/articles/360018834314-Stori... ) and it seems the blogger must opt-in into it, so maybe the downvoters also had same impression as me.
Still weird that some person wants to promote some conference and fine volunteers and puts it under a paywall.
1) I'm going to guess that the count of independent bloggers whose primary income derives from their blog is, at best, in the three digit range.
2) I can't speak for others, but paywalls have convinced me to give money to zero entities. I do subscribe to a couple, but making me delete my cookies was not their sales tactic.
3) Completely aside from all that, none of this says anything about HN, and is worth discussion. If I post a link to one of the sites I subscribe to with a larger price tag, where's the line? At what point does it damage conversation? I wonder if, on average, there's less discussion of paywalled links, or if the participant pool is different.
Just as one datapoint, I think Google News should not include WSJ links in the default feed. It is nothing but an ad, which is ironic considering the rest of news.google.com is ad-free.
They allow a couple of articles per month before you hit a paywall. It's frustrating because there happens to be a ton of interesting content (esp. machine learning-related) on Medium and the exact same content could be in a Wordpress or other blog (with a Medium-like theme if you'd like that), and it takes a total of 5 minutes to set up a blog.
Statistical machine learning is one of the fields where diversity really does matter. A CPU from Silicon Valley works just as well in Africa. But models trained in the West and with hyperparameters tweaked by western people will contain all kinds of biases that make it dangerous even in the west and downright lunatic in a very different culture.
I can't imagine a self-driving car managing to drive on South African (or many other African) roads. It seems like these cars are trained to drive in countries with different types of driver (and pedestrian) behaviour, and probably wouldn't generalise well to places with little regard for rules of the road. Mini-bus taxis here pretty much do whatever they want and are far from rational in their decision making...
I wouldn't be surprised if Africa eventually surpassed the West in AI. Often it's easier for countries and companies with less existing technological infrastructure (or baggage from the previous dominant paradigm) to ramp up on something new.
Knowledge workers aren't the same as physical infrastructure. Where you see the developing world surpass the West are things such as payment systems, cell networks, trains etc. Physical things that the West has invested in previous, where the developing world isn't hampered by the momentum of existing systems.
I would be very surprised. ML is basically advanced applied mathematics and requires a high level of technical literacy. Africa in general lags the world in most fields of STEM. There are plenty of examples where some developing country has better cell networks than the US but fails to come close to Western cutting edge research.
This is sometimes called "leapfrogging" and one example was cell phone technology (when countries that came later to the game adopted newer and better standards).
Surely he must mean implementation in the overall infrastructure. From my time in America is seems that overhauling systems is not popular policy at all. Unfortunately as an African myself there are larger issues than AI that need to be solved before that can even be in the periphery of something near that statement.
- Kenya has a dominant electronic currency, M-Pesa
- South Korea rapidly developed into a globally competitive developed country, despite being poor in the middle of the 20th century (same applies to Japan, just slightly less so).
Yeah and they will use all the deep learning research, probably open source tools like TensorFlow, and then deploy to AWS? I think what's happening is that SV is democratizing IT tools for some years now and has been expanding globally. Bit weird to frame it as a competitor when most likely it's probably a child of silicon valley, and cannot really compete unless they come up with a COMPLETELY NEW infrastructure and AI technology from the ground up, which is possible but obviously not the case.
It's good to get away from the Silicon Valley mold. Sure it has produced many great successes, but it's only one method of going about things, with various trade-offs and failure modes. There's no reason for the model of seed-> angel-> venture-> IPO has to be repeated everywhere, a robust ecosystem of technological advancement should have more diversity than that.
Offtopic: This article contains four pictures that contribute nothing to the content and weigh around 10MB. The first one takes up a whole screen of space. Why do people do this?
It's probably to do with google's algo that favours a certain content-to-image ratio. Medium gives the author access to easily embeddable images that often have nothing to do with the article just so the article's not just one giant block of text.
I would have loved to have seem some photos of this industry out of Africa though.
Ok good, but why do these pictures have to be 2-3MB each? Each one of them would have taken an hour to download just a few years ago (or today, if you live in a less well developed area of the world).
I wonder if there is a way browsers could preemptively “compress“ (I doubt that’s the correct terminology) prior to downloading? Like somehow grab every nth bit, combine, render and then allow for further requests to refine at user discretion. Wonder if anyone has ever tried something like that.
I don't think that's even possible with the byte-ordering of a lot of the formats we use today.
Like the pass extraction of PNG [0] would make it somewhat possible, but you'd need to be dropping packets which you might as well receive. I don't think you'd be able to break after the first pass.
JPEG is probably worse.
However, the super-modern formats that are still gaining popularity, like WebP absolutely do support this. At the core of WebP is block prediction, so a low-bandwidth application might actually assume the values of certain blocks so it doesn't need to fetch them.
I was downloading MP3s to my MP3 player with a 16MB san over ISDN circa 2000. For most people in the US it hasn’t taken hours to download an MP3 for over a decade maybe 2.
I'd rephrase that into skimmers are often driven away by a block of text while readers are driven away by page-filling images. There is an optimum between these two which attracts people from both sides of the divide - a few targeted images can enhance a text and make it look less imposing to skimmers.
And in all fairness, this isn't just a web thing. Most newspapers have historically used all manner of graphical techniques in addition to images to break up text blocks such as subheads, callouts, graphics, lines, etc. (Although, also to be fair, historically newspapers also tended to try to include images that were directly relevant to the story rather than picking random stock imagery.)
Have you ever read a “magazine”? These were pre-internet things that were printed on paper, and included text and images. Often these images were illustrations, rather than photographs. They break up the text walls, attract the eye when you’re flipping through deciding which piece to read, and create a mood.
As to why the images are so huge, well, nobody gives two shits about download size any more in a world where you’re also probably pulling down multiple megabytes of ad images and JavaScript because every damn webpage has to be an app even if it’s purely static content nowadays.
Because even with 100+ million of VC money [1], Medium couldn't be bothered to pay for a CDN that automatically optimizes images and/or optimize the image on their own after the user uploads it.
While the practical notebooks we used in 2018 were already amazing, my colleagues and I worked hard on expanding and improving them for the 2019 Indaba. In case anyone wants to check them out, here's a full list of the practicals (the links go to Google Colab notebooks)
1a: Machine Learning Fundamentals, http://bit.ly/2nfeu4W
1b: Build your own TensorFlow, http://bit.ly/2obD1rK
2a: Deep Feedforward Networks, http://bit.ly/2oQSjSQ
2b: Optimization for Deep Learning, http://bit.ly/2nqkyHB
3a: Convolutional Networks, http://bit.ly/2nqdQRX
3b: Deep Generative Models, http://bit.ly/2nnkisU
4a: Recurrent Neural Networks, http://bit.ly/2ncgYkx
4b: Reinforcement Learning, http://bit.ly/2oO4lMK
They're also available on the official github page: https://github.com/deep-learning-indaba
We'd love any feedback on the practicals to improve them for next year!
If you have some ML expertise and would like to volunteer to be a tutor at next year's Indaba, which will be held in Tunisia in late August 2020, drop me a mail at contact (at) taliesin.ai. I also encourage anyone to apply to attend as a student!