I read up to “of this post.”. Took me way too long to realize pseudo-inputs are input sets/distributions, not particular inputs. I’m guessing the argmax is supposed to be a max. Why do you split P(α(x) and C(M,x)) into P(α(x))*P(C(M,x)|α(x))?
Good catch! Also, I generally think of pseudo-inputs as predicates, not particular inputs or sets of inputs (though of course a predicate defines a set of inputs). And as for the reason for the split, see the first section in “Other approaches” (the basic idea is that the split lets us have an adversary, which could be useful for a bunch of reasons).
I read up to “of this post.”. Took me way too long to realize pseudo-inputs are input sets/distributions, not particular inputs. I’m guessing the argmax is supposed to be a max. Why do you split P(α(x) and C(M,x)) into P(α(x))*P(C(M,x)|α(x))?
Good catch! Also, I generally think of pseudo-inputs as predicates, not particular inputs or sets of inputs (though of course a predicate defines a set of inputs). And as for the reason for the split, see the first section in “Other approaches” (the basic idea is that the split lets us have an adversary, which could be useful for a bunch of reasons).