Jonathan Hersh - Applying Deep Learning to Satellite Imgs to Est Violence in Syria & Poverty in MX

[@Greyz174]

- 2:45 - Why we don't collect good poverty data - 4:30 - Talking about micro-satellites - 5:10 - Talking about using computer vision to extract cars, building, road data from sat images - 6:45 - Introduce the use of transfer learning - 9:26 - Using off the shelf data is for suckers - 10:25 - Incompleteness of violence data - 11:24 - Use computer vision to detect bomb damage in Syria - 15:35 - Note that data even from sat images is imperfect (but improving) - 15:53 - Using data to predict refugee flows, analyze reporting bias etc. - 16:15 - Dynamic Models > Static Models I was bookmarking time points in the video because I'm using this for a school project. Since it's just a copy/paste away, thought I'd post a rough outline in the comments. Note that this is what I care about, not a complete outline. But in case it does anyone any favors, thought it might help.