Computer Vision News - December 2021

15 Multimodal Single-Cell Data Integration were intuitive for someone with a machine learning background, while introducing some of the background knowledge to explain what people are looking at and what the different tools are. The idea for the competition formed a couple of years ago when a paper called ’Eleven grand challenges in single-cell data science’ came out. “ I really liked thepaper, ”Daniel tellsus. “ One of our co-organizers was a co-author of it, so there’s no bashing here, but it was very generic andnot formalized, so itwasn’t clear what the problem was you needed to find a solution for. In the field of mathematics there are well-defined problems for which we don’t have effective solutions. How do you mathematically quantify what you would like people to solve? ” Partofthegoalofthiseffortwastotakesome of that formalization and mathematical rigor and apply it to something which might just be called a challenge. However, it is more than that – it is a huge open problem that takes a lot of work to solve, so the team wanted to reflect that in the name of protein in individual cells, at this 100,000 million cells scale. One of the nice things is multimodal data has a built-in ground truth because you can say I measured one part of this system and then in a paired manner I measured this other aspect. ” This throws up questions such as: If I only expose you to one, can you predict the other? What if I give you many sets of both, and I want you to match them? What if I have a set of annotations I like about the data and I want you to learn a joint embedding into a lower-dimensional space where certain properties are preserved? The goal of the competition is to introduce some of these tasks for the first time for multimodal single-cell data, and to incentivize broad-scale competition from statisticians, computational biologists, and machine learning practitioners to bring these communities together in a way that has not been done before. The team have aimed to provide as straightforward a task as possible to enable people who do not necessarily have a life sciences background to participate. They came upwith tasks that