English

Actively Learning Halfspaces without Synthetic Data

Data Structures and Algorithms 2025-09-26 v1 Machine Learning

Abstract

In the classic point location problem, one is given an arbitrary dataset XRdX \subset \mathbb{R}^d of nn points with query access to an unknown halfspace f:Rd{0,1}f : \mathbb{R}^d \to \{0,1\}, and the goal is to learn the label of every point in XX. This problem is extremely well-studied and a nearly-optimal O~(dlogn)\widetilde{O}(d \log n) query algorithm is known due to Hopkins-Kane-Lovett-Mahajan (FOCS 2020). However, their algorithm is granted the power to query arbitrary points outside of XX (point synthesis), and in fact without this power there is an Ω(n)\Omega(n) query lower bound due to Dasgupta (NeurIPS 2004). In this work our goal is to design efficient algorithms for learning halfspaces without point synthesis. To circumvent the Ω(n)\Omega(n) lower bound, we consider learning halfspaces whose normal vectors come from a set of size DD, and show tight bounds of Θ(D+logn)\Theta(D + \log n). As a corollary, we obtain an optimal O(d+logn)O(d + \log n) query deterministic learner for axis-aligned halfspaces, closing a previous gap of O(dlogn)O(d \log n) vs. Ω(d+logn)\Omega(d + \log n). In fact, our algorithm solves the more general problem of learning a Boolean function ff over nn elements which is monotone under at least one of DD provided orderings. Our technical insight is to exploit the structure in these orderings to perform a binary search in parallel rather than considering each ordering sequentially, and we believe our approach may be of broader interest. Furthermore, we use our exact learning algorithm to obtain nearly optimal algorithms for PAC-learning. We show that O(min(D+log(1/ε),1/ε)logD)O(\min(D + \log(1/\varepsilon), 1/\varepsilon) \cdot \log D) queries suffice to learn ff within error ε\varepsilon, even in a setting when ff can be adversarially corrupted on a cεc\varepsilon-fraction of points, for a sufficiently small constant cc. This bound is optimal up to a logD\log D factor, including in the realizable setting.

Keywords

Cite

@article{arxiv.2509.20848,
  title  = {Actively Learning Halfspaces without Synthetic Data},
  author = {Hadley Black and Kasper Green Larsen and Arya Mazumdar and Barna Saha and Geelon So},
  journal= {arXiv preprint arXiv:2509.20848},
  year   = {2025}
}
R2 v1 2026-07-01T05:55:30.868Z