Hi all, I’m Dave!
Very happy to be joining everyone for this next version of the course. I’m currently consulting for the World Bank / Global Fund for Disaster Reduction and Recovery on geospatial ML for disaster risk management. Small plug: I’m an organizer of the Open Cities AI Challenge, an active competition (until 3/16!) with a novel, accessible dataset of building footprints and drone imagery spanning 10 African cities and $15K in total prizes for semantic segmentation and responsible AI ideas for improving disaster resilience. I also work independently on AI for climate change projects and plan to put more emphasis there going forward.
My deep learning journey is essentially a fastai journey, having taken almost every remote version of the course with a happy exception of doing part 2 (2018 version) in-person. I don’t have a coding background (formal education is in bio/medicine/business) so these courses and community have doubled as excellent resources on python programming and software dev. This is one of the most useful educational experiences I’ve engaged in (with more than enough formal schooling to compare against!). And it’s through my learning projects done during the courses (i.e. coconut tree detector, Zanzibar building segmentation with drone imagery tutorial) that helped me get to my current work. It’s a testament to the quality and depth of the teaching that I learn so much and still feel like I barely scratch the surface in every iteration.
As a stretch learning goal for this course, I would love to work with anyone interested on creating a geospatial submodule for fastai2 in the style of fastai2.medical.imaging. This may be too time-ambitious (at least for me as a recently new parent) but as a minimum, I’d like to put together some demo notebooks showing how to take in some useful open geospatial datasets and achieve good-to-great results on common geoML applications like land cover mapping with the power of fastai2.
Looking forward!
Dave