|
|
|
|
|
|
|
|
|
|
|
n |
Choosing
queries that reduce the expected loss of our
|
|
estimate as much
as possible
|
|
|
n |
But… are we
converging to the “correct” distribution?
|
|
|
n |
Theorem: Our
method is consistent
|
|
|
|
Ž |
q converges to true q that is generating the data
|
|
|
Ž |
Reducing loss as fast as possible.
|
|
|
And converging to the right solution.
|
|