Properly learning monotone functions via local reconstruction

TL;DR

This paper introduces a nearly optimal proper learning algorithm for monotone Boolean functions that is robust to noise, using local computation techniques for sorting on posets, with applications to tolerant testing.

Contribution

It presents the first proper learning algorithm for monotone functions with subexponential time, leveraging local sorting algorithms and extending to tolerant testing on posets.

Findings

Achieves $2^{ ilde{O}(rac{ oot{n} ext{}}{ ext{epsilon}})}$-time proper learning algorithm.

Provides a local computation algorithm for sorting binary labels on posets.

Extends tolerant testing to general posets with improved accuracy bounds.

Abstract

We give a $2^{\tilde{O}(\sqrt{n}/\epsilon)}$-time algorithm for properly learning monotone Boolean functions under the uniform distribution over $\{0,1\}^n$. Our algorithm is robust to adversarial label noise and has a running time nearly matching that of the state-of-the-art improper learning algorithm of Bshouty and Tamon (JACM '96) and an information-theoretic lower bound of Blais et al (RANDOM '15). Prior to this work, no proper learning algorithm with running time smaller than $2^{\Omega(n)}$ was known to exist. The core of our proper learner is a \emph{local computation algorithm} for sorting binary labels on a poset. Our algorithm is built on a body of work on distributed greedy graph algorithms; specifically we rely on a recent work of Ghaffari (FOCS'22), which gives an efficient algorithm for computing maximal matchings in a graph in the LCA model of Rubinfeld et al and Alon et al (ICS'11, SODA'12). The applications of our local sorting algorithm extend beyond learning on the Boolean cube: we also give a tolerant tester for Boolean functions over general posets that distinguishes functions that are $\epsilon/3$-close to monotone from those that are $\epsilon$-far. Previous tolerant testers for the Boolean cube only distinguished between $\epsilon/\Omega(\sqrt{n})$-close and $\epsilon$-far.