My understanding of this paper/space is patchy-at-best, but I am I right in thinking that this could only ever work for models where both the outcome-variable and the predictors are entirely categorical, and the input data is always encrypted with the same key?
(E.g. so that the value 'male' in the 'sex' column has the same ciphertext across subjects/observations).
If this is the case, then couldn't a standard logit model be used? (It needn't know what it's categories 'mean', either x or y).
These properties would be fulfilled by, e.g. the MNIST example in the paper (and it wouldn't take 17 hours to train!). Does anyone know where the technique in the paper would work but the one described above would fail?
(E.g. so that the value 'male' in the 'sex' column has the same ciphertext across subjects/observations).
If this is the case, then couldn't a standard logit model be used? (It needn't know what it's categories 'mean', either x or y).
These properties would be fulfilled by, e.g. the MNIST example in the paper (and it wouldn't take 17 hours to train!). Does anyone know where the technique in the paper would work but the one described above would fail?