[License-discuss] License on neural net data?

Dr. David Alan Gilbert gilbertd at treblig.org
Sun Jun 19 19:31:26 UTC 2016


Hi,
   I wondered if there are any licenses that are appropriate for the
weight data in neural-network/learning systems. (It's mostly
an idle thought rather than something I've got a project on).

   For example, if I were to train a voice-recognition system on
a mix of source audio under a bunch of different licenses, what can I license
the trained system under?

   It seems in some ways some systems can be seen as lossy compression
systems, e.g. (1) - so those are on the edge of having a lot of the original
source, but that doesn't feel like it should be true of something trained
for voice recognition say - I bet the border between the two cases will
get some interesting cases.

   In the case of voice-recognition I was thinking of trying to train
things on speeches given at (open-source) conferences, which is what got
me thinking about it.

   There are probably some fun other cases, like if someone starts with
your model and then trains it a lot more, what are the terms they're
allowed to distribute their new model under.

Dave
(just gently thinking about AI stuff)

1: https://medium.com/@Terrybroad/autoencoding-blade-runner-88941213abbe#.uhemm6z4u

-- 
 -----Open up your eyes, open up your mind, open up your code -------   
/ Dr. David Alan Gilbert    |       Running GNU/Linux       | Happy  \ 
\        dave @ treblig.org |                               | In Hex /
 \ _________________________|_____ http://www.treblig.org   |_______/



More information about the License-discuss mailing list