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?
That’s actually an interesting question - I would lean towards it being a sufficiently reasonable transformation but I’m obviously not a lawyer.<p>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.<p>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.<p>Again not saying who is right (in terms of the law), just think aloud :)