Ask HN: IS a ML model derivative of the data it was trained on?

16 pointssource997y ago4 comments

Curious on what people think about this. If I use an open source dataset to train a model is my model bound by the same license as the dataset? or do I have more control over my own model?

4 comments

4 comments · 2 top-level

olliej7y ago· 2 in thread

That’s actually an interesting question - I would lean towards it being a sufficiently reasonable transformation but I’m obviously not a lawyer.

That said more or less all modern ML is just a glorified statistical model. That it allows a non-linear model is the bit that makes it useful vs. basics like averages and deviations.

The reason I say that is because (in the US), the NFL and MLB, etc have frequently stated that they believe that even the statistics (not just the raw data) are their property. If they consider linear statistical values to be property I’m sure they would think the same about those properties that make up the weights in a NN.

Again not saying who is right (in terms of the law), just think aloud :)

setr7y ago

In some other thread recently and iirc, it was being claimed with some confidence that the statistics claim was determined not to be the property of the NFL/MLB, by some court case, as you can't claim rights to facts about the world. You can claim rights to representations of facts, such as a visualization, but not to the facts themselves, and their arbitrary derivatives.

Which, at least, sounds like a sane argument to me; it would be absurd to think that I couldn't take notes on a game being shown on television, and try verifying whatever numbers the NFL report for a game/series.

olliej7y ago

Sounds sane to me, but occasionally that doesn’t reflect what courts think :)

kazinator7y ago

Also: is your education derivative of all those textbooks?

j / k navigate · click thread line to collapse