I’ve been reading Range by David Epstein and damn do I have a lot of notes.
Hopefully fuel for some future blog posts. But, for now, I haven’t been able to get over this section in a chapter on non-experts solving problems that bedevilled experts.
“Shubin Dai, who lives in Changsha, China, was the top-ranked Kaggle solver in the world as of this writing, out of more than forty thousand contributors. His day job is leading a team that processes data for banks, but Kaggle competitions gave him an opportunity to dabble in machine learning. His favorite problems involve human health or nature conservation, like a competition in which he won $30,000 by wielding satellite imagery to distinguish human-caused from natural forest loss in the Amazon. Dai was asked, for a Kaggle blog post, how important domain expertise is for winning competitions. “To be frank, I don’t think we can benefit from domain expertise too much… It’s very hard to win a competition just by using [well-known] methods,” he replied. “We need more creative solutions.””
This is referring to Kaggle, a site I haven’t yet found the bravery to crack with my new coding skills 😅
Anyway, it comes amid an exploration of problem solving, and how experts can be trapped by their domain knowledge, falling back into familiar patterns that haven’t worked. Meanwhile the answer, as in the case of some of these kaggle competitions, ie often from somewhere else completely.
““The people who win a Kaggle health competition have no medical training, no biology training, and they’re also often not real machine learning experts,” Pedro Domingos, a computer science professor and machine learning researcher, told me. “Knowledge is a double-edged sword. It allows you to do some things, but it also makes you blind to other things that you could do.””
As always my emphasis