Core ML framework why baked in the bundle

Question

In my limited knowledge of the new Core ML framework, I understand that the model is build in to the bundle.

I was sure that ML and models evolve over time, collecting more data to furthere evolve the model. If that is true, why is the Core ML framework then using "static" models?

I can see that for speed needed for computer vision(object recognition etc.), but the model should be able to evolve reading from http something - right?

score 4 · Accepted Answer · edited Jul 05 '17 at 22:31

4

CoreML is not made to train the model on the mobile device, but to run a trained model in the app.

It's still true that you might improve your model over time and might want to be able to update it without having to submit a new app.

There has been some discussion on that: https://stackoverflow.com/a/44463680/2054629 and I don't have a real answer on why Apple didn't made easy to update a model as long as it takes the same input and has the same output types.

edited Jul 05 '17 at 22:31

Alex Brown

41,819
10
94
108

answered Jul 01 '17 at 04:30

Guig

9,891
7
64
126

Maybe they figured that if you wanted to update the model, you'd update the whole app. – Matthijs Hollemans Jul 01 '17 at 15:47
Yeah but that's not great. You can also have an app that uses few models among a large choice and wants to download them as needed. – Guig Jul 01 '17 at 18:36
@Guig mainly because nobody asked that question out loud yet. You are correct that there's some discussion in https://stackoverflow.com/a/44463680/2054629 but since that's not that OP's question, it would be inappropriate to answer it there. Please feel free to ask that question. – Alex Brown Jul 05 '17 at 22:30

Core ML framework why baked in the bundle

1 Answers1