My position on generative AI
Please stop shoving it down all our throats.
A reverse Cantelli inequality
It's the Cantelli inequality, but reversed.
Topological properties of adversarial examples
In one of the earliest papers on adversarial examples for deep neural networks, it was suggested that adversarial examples may be "dense" in the input space in the same way rational numbers are dense in the reals. While I cannot prove or disprove this conjecture, I show here that it would be big if true.
Fun with differentials
I investigate the question of whether, for any $$n$$, there exist functions $$f$$ whose $$n$$th order derivative $$f^{(n)}$$ is again equal to $$f$$.
A simple upper bound on adversarial robustness
I prove a simple upper bound on the robustness of linear classifiers in the $$\ell_2$$ threat model.