0
$\begingroup$

I'm starting my machine learning study and I'm trying to figure out a simple question:

Let's say that I have two models, one that recognizes cats, and another one that recognize dogs.

Now I have a camera and I want to recognize both cats and dogs using my models.

Obviously, I don't want to create a third model to recognize both, having all the work to label each animal, so, is there a method to "merge" both models into one?

I'm asking this question because I want to understand one thing: why ML engineers don't share their models so then we can create aggregated models? For example, if a person A has a model that classifies people and a person B has a model for animals. Why they can't just share their models so then each one will have a more powerful model without needing to re-train everything?

I'm sorry if this is a too basic question, but I didn't see on Google any clear explanation. Thanks.

$\endgroup$

2 Answers 2

1
$\begingroup$

I believe that I've found an answer here and here.

$\endgroup$
1
1
$\begingroup$

You absolutely can do this, and it is very often done in practice. Many models are publicly available, and since training takes a lot of time, a lot of ML engineers will use pretrained models to begin tackling their problem.

And your intuition is right, that re-training for the "combined" scenario would be a waste of time.

If you want some more reading, look into ensemble methods and transfer learning.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.