English

Superpolynomial Lower Bounds for Learning Monotone Classes

Data Structures and Algorithms 2023-01-31 v2

Abstract

Koch, Strassle, and Tan [SODA 2023], show that, under the randomized exponential time hypothesis, there is no distribution-free PAC-learning algorithm that runs in time nO~(loglogs)n^{\tilde O(\log\log s)} for the classes of nn-variable size-ss DNF, size-ss Decision Tree, and logs\log s-Junta by DNF (that returns a DNF hypothesis). Assuming a natural conjecture on the hardness of set cover, they give the lower bound nΩ(logs)n^{\Omega(\log s)}. This matches the best known upper bound for nn-variable size-ss Decision Tree, and logs\log s-Junta. In this paper, we give the same lower bounds for PAC-learning of nn-variable size-ss Monotone DNF, size-ss Monotone Decision Tree, and Monotone logs\log s-Junta by~DNF. This solves the open problem proposed by Koch, Strassle, and Tan and subsumes the above results. The lower bound holds, even if the learner knows the distribution, can draw a sample according to the distribution in polynomial time, and can compute the target function on all the points of the support of the distribution in polynomial time.

Keywords

Cite

@article{arxiv.2301.08486,
  title  = {Superpolynomial Lower Bounds for Learning Monotone Classes},
  author = {Nader H. Bshouty},
  journal= {arXiv preprint arXiv:2301.08486},
  year   = {2023}
}
R2 v1 2026-06-28T08:16:03.269Z