The Deep Learning Indaba started from a blunt observation. At the major machine learning conferences, African representation was close to nonexistent, not for want of talent but for want of the connective tissue that makes a research community function: people who know each other, mentors who will read a draft, a room where you can present something unfinished and get useful pushback. Conferences are an expensive and visa-dependent way to get that if you are outside the usual geographies. So they built their own, annually, and it has grown into the central meeting point for machine learning on the continent.
The 2026 edition ran in August in Lagos, hosted at Pan-Atlantic University, with applications having closed the previous March. That structure is worth understanding if you want to attend: this is not a course you enrol in, it is an event you apply to, and the deadline sits roughly five months before the week itself. Students, early career researchers, faculty, industry and public sector people are all eligible, and everyone attending in person goes through selection regardless of whether they are paying a registration fee. There is a virtual track for talks and discussions.
What makes it interesting for people who will never attend is that the teaching material gets published. The practicals, which are Jupyter notebooks written for the week, go up openly, and they are good. Written by researchers for an audience of capable people who may not have had access to a well-resourced ML programme, they tend to be self-contained, carefully scaffolded and honest about what is hard. Over the years these have covered the standard foundations plus topics that reflect the community's own priorities, including work on African languages and low-resource NLP, which is genuinely undersupplied in mainstream curricula.
If your interest is multilingual or low-resource work, this is one of the better material sources anywhere, and that is not a charitable framing, it is a statement about where the expertise actually sits. The honest limitation is the same one every summer school has. A week is a week. You cannot learn deep learning in five days and nobody involved claims you can.
What a week can do is orient you, show you what is possible, and put you in a room with people who are two or five years further along, and the Indaba is explicitly built for that third thing. Its design emphasis is on relationships, with mentorship structures, satellite IndabaX events run in individual countries through the year, and an alumni network that keeps working after the week ends. Judged as a course it is thin. Judged as community infrastructure it is one of the more effective things anyone in this field has built.
The practical advice depends on who you are. If you are in the African research community, apply, and apply early, and note the March deadline now. If you are elsewhere, download the practicals and work through the ones relevant to you, because they are free, openly licensed and better written than most paid alternatives. If you run any kind of technical education programme, look at how the IndabaX model devolves teaching into local events, because it works and hardly anybody copies it.
Four point four.