I wonder if you could apply the same homomorphic encryption to ML model training as well, so that you can fully train an algorithm without ever being able to ac

I wonder if you could apply the same homomorphic encryption to ML model training as well, so that you can fully train an algorithm without ever being able to actually “see” the training data.