The preprint will be published in about a month, in the meantime you can check out the project page here: lnkd.in/eYFDYp2f (5/5)

Luckily, I found that addressing the bias doesn't have to be complicated and does not have to compromise accuracy. I presented this work at the MACLEAN workshop at ECML-PKDD +very happy to receive the Best Paper Award for our research on mitigating representation bias caused by missing pixels. (4/5)

One of the most important but also tricky parts of building datasets is sampling/selecting instances, but even with support from domain experts, you can accidentally introduce bias or confounders. (3/5)

It's common data science and machine learning knowledge that data cleaning is at least half of the work, but it's not until I started my PhD that I understood the extent of how much care you need to put in curating a dataset, especially with real data. (2/5)

We're also calling for pitches on environmental/wildlife stories and looking for new partnerships with environment-focused organisations. If you have an open source based investigation that might be a good fit for us, please get in touch. Details are here: www.bellingcat.com/uncategorize...