Re: legal questions regarding machine learning models
2009/5/29 Mark Weyer <weyer@informatik.hu-berlin.de>:
> Am I missing something? I would think that even if in all jurisdictions
> the font is non-copyrightable, that still would not imply DFSG-freeness,
> only that it is fit for non-free.
>
> Best regards,
>
> Mark Weyer
That's what I thought as well cause source is not available in
preffered form of modification. But imagine if noone in debian knew
that this raster font was generated from something else, then it would
DFSG-free.
So just expanding on that. DFSG source requirement is concluded by
judging each time what is source. And this is biased sometimes as we
see in this example. The model (I presume in somekind of human or
machine parsable format) if distributed under free license does allow
to view all parameters and tweak them. (It will be wrong from
scientific point of view but just for fun "what if twist this nob
without actually parsing any data for many days in a row") I would
argue that is still source. Now imagine we have many of these models
and someone writes a hyper-model simulator which takes all of these
models and make a new one based on some cunning statistical
processing....... what will be source then? all input data to all
models with all the generation parameters and environments? (i bet
some of them use urandam as well)???? or will the models become source
for the hyper-model?
--
With best regards
Dmitrijs Ledkovs (for short Dima),
Ледков Дмитрий Юрьевич
Reply to: