Computer Vision News - December 2021

16 NeurIPS Challenge around the world with 37 people involved with generating the data, processing it, analyzing it, andmaking sure itwas properly formatted. Part of what makes this so special is that the effort brings together wet lab biologists who have no experience with machine learning and machine learning practitioners who have little prior experience with single cell biology. Just generating a benchmarking dataset is a landmark effort. How did they manage to get so many people motivated towards a common goal? “ A lot of people just love science and they’re really interested in biology! ” Daniel responds. “ People are just innately curious about what is going on with this brand- new kind of data and they all understand the value of having these methods properly benchmarked. ” The biggest barrier to this work has been the funding. The team managed to secure financial sponsorship from both Cellarity and the Chan Zuckerberg Initiative . “ It’d be remiss if I didn’t acknowledge that the initiative: Open Problems. Daniel says this work excites him because he uses these kinds of methods every day, and it can be frustrating trying to figure out what tool to use for a given task. Also, he just loves bringing people together. “ Academia can get very competitive – it’s funny I’m saying academia is competitive and one of the ways I’m going to solve that is through a competition! ” he laughs. “ But there’s a lot of good-natured effort by people who want to solve these problems and want to know if they’re doing well or not compared to some other tool. It makes it a lot easier if you’ve already come up with the task you’re trying to solve because then people have a clear understanding of how well they’re performing and in what context. When are they better than another tool? When are they not? There’s a lot of interest from many kinds of people to want to come together. ” Generating the dataset for this competition was a huge team effort that took place across four different academic institutions

RkJQdWJsaXNoZXIy NTc3NzU=