← Back to the course page | Compare: distinguishers that see the model's output →

Animation — distinguishers that depend only on x

The BOOST procedure with ε = 0.05 and the class of tests F={1, x, x², x³, x cos x}, each depending only on x. At each step the orange bar is the test selected for the update (the largest violator); the procedure halts when every test is inside the ±ε band.

Boosting animation with x-only tests, per-test error, and L2 error — Left: target p^* and model *p_t*. Middle: the indistinguishability error of each test, Right: the l2 distance between the truth and the model

The companion animation runs the same procedure with a class that also contains tests of the model's output p(x), and reaches a lower distance.