[OSM-talk] [OSM-dev] Update from HOT on machine learning in the Tasking Manager

[Christoph Hormann]

On Monday 17 September 2018, Nate Smith wrote:
>
> Read the full post here:
> https://www.hotosm.org/updates/integrating-machine-learning-into-the-
>tasking-manager/

That looks refreshingly different in focus looking more at data quality 
assurance and validation and less at automated data production.  I look 
forward to seeing more practical results in that direction.

One specific hint about this:  For validation and QA purposes AI 
techniques are often inefficient and noisy/error-prone and traditional 
image analysis can sometimes be at least as valuable and reliable with 
less ressource requirements.  Because in validation you have a priori 
facts/claims you want to validate, you need to answer questions 
like "is this building geometry accurate" and not the tasks AI 
techniques are typically advertised for like "find and trace all 
buildings in this area".

There are of course also QA tasks where AI can be more useful for like 
the typical needle-in-a-haystack problems.

-- 
Christoph Hormann
http://www.imagico.de/