All of frisby's Comments + Replies

A couple small notes / clarifications about the results in this post:

  • Theorem 1 could be equivalently stated "For any constant , the family of size- circuits is Verifiably Acceptable". (If  is parameterized by descriptions that can be evaluated in polynomial time, then you can wlog think of  as a subfamily of polynomial-sized circuits.)
  • It's worth emphasizing that while  (the location of the backdoor) is chosen uniformly random here, both  (the function we'd like to distill, but can only access as a b
... (read more)

Confirming that efficiently finding a small circuit (you don't actually need further restrictions than size) based on its values on a fixed collection of test inputs is known to imply  --- see this paper.

This is a fact worth knowing and a lucid explanation --- thanks for writing this! 

I know it's not the main point of the post, but I found myself a little lost when you talked about complexity theory; I would be interested to hear more details. In particular:

  • When you say

           what are the definitions of these classes? Are these decision problems in the usual sense, or is this a statement about learning theory? I haven't been able to either track down a def... (read more)