Preparing: ML-Policy: Debian Policy Enhancement Proposal
Hi Debian Developers,
Deep learning is not far from you. Maybe next day
it will land onto your computer through a fancy
code completion plugin or something alike.
Again I renamed the "DL-Policy" into ML-Policy.
Please help me move a step further on standarlizing
the ML-Policy, before really filing a BUG against
the Debian Policy. My goal is to get it merged in
the end (why shouldn't I?).
I've moved the explanations and all the introductory
words from the main document to another file `faq.rst`
(explanations are still not organized after reconstruction).
As a result, the main document only contains the draft
for the proposed policy section, very concise:
https://salsa.debian.org/lumin/ml-policy/blob/master/ML-Policy.pdf
https://salsa.debian.org/lumin/ml-policy/blob/master/ML-Policy.html
The source is in restructuredText format, but gitlab
doesn't render rst. PDF is my recommended format for
reading.
Other changes:
1. Removed nonsense from the definitions and the terms.
Now the definitions should be much clearer.
2. A part of "ToxicCandy" model has been split into
a new category "Sourceless model". This new
category is for models that doesn't cause
serious problem even if misbehaved. See document
for the precise definition.
Any suggestion, comment will be appreciated, as usual.
Dear developers, please tell me more cases to be
studied, especially if your package ships pre-trained
machine learning models (especially neural network)
if you aren't sure how to classify it. Currently I
don't have much examples to study and verify my
proposal:
https://salsa.debian.org/lumin/ml-policy/blob/master/case-study.md
Please send mails to me or submit a issue here:
https://salsa.debian.org/lumin/ml-policy/issues
I'll review all the cases you provide.
After reviewing enough amount of cases, I'll
start a standard RFC.
Apart from the policy proposal, there are still
some ambiguous problems when a deep learning
software is licensed under GPL-*. I think I should
assemble a list of such problems in the future,
and maybe contact FSF for help.
Let's prepare for the era of AI?
Reply to: