Seeking advice on how to approach a situation where we have to train the same simple model (not NNs) for each individual user and their content which must remain private - this means I would theoretically have 150k different models to deploy/operate in parallel (possibly a nightmare).
Is it possible in TensorFlow to train/freeze these individual models completely separately, and then load + combine them into one larger model for saving at deployment time (as no training is required) that takes a userId as an additional input.
If not, is there a typical approach to this problem?