[OSM-dev] Loc2Vec

Frederik Ramm frederik at remote.org
Mon Jun 18 18:00:01 UTC 2018


On 06/18/2018 07:30 PM, Christoph Hormann wrote:
> I think it is probably possible to create scenarios where all three 
> viewpoints (the two above and the 'no copyright issues involved' case) 
> are plausible interpretations.

I can certainly construct a primitive "machine learning" machine that
can be "trained" with OSM data and that will nicely respond with that
very OSM data very time I set it to work on the same geographical area ;)

I have a hunch that in order to disprove that your machine learning
installation and all it produces is a derived work, you would at least
have to lay open the algorithms and the training data used.

If someone has a proprietary machine and trains it with OSM data, then
my default assumption would be that everything the machine ever outputs
is derived from OSM, because how will I know that the machine doesn't
simply store everything it sees during learning?


Frederik Ramm  ##  eMail frederik at remote.org  ##  N49°00'09" E008°23'33"

More information about the dev mailing list