ResearchTrend.AI
  • Papers
  • Communities
  • Events
  • Blog
  • Pricing
Papers
Communities
Social Events
Terms and Conditions
Pricing
Parameter LabParameter LabTwitterGitHubLinkedInBlueskyYoutube

© 2025 ResearchTrend.AI, All rights reserved.

  1. Home
  2. Papers
  3. 2002.04840
19
42

Efficient active learning of sparse halfspaces with arbitrary bounded noise

12 February 2020
Chicheng Zhang
Jie Shen
Pranjal Awasthi
ArXivPDFHTML
Abstract

We study active learning of homogeneous sss-sparse halfspaces in Rd\mathbb{R}^dRd under the setting where the unlabeled data distribution is isotropic log-concave and each label is flipped with probability at most η\etaη for a parameter η∈[0,12)\eta \in \big[0, \frac12\big)η∈[0,21​), known as the bounded noise. Even in the presence of mild label noise, i.e. η\etaη is a small constant, this is a challenging problem and only recently have label complexity bounds of the form O~(s⋅polylog(d,1ϵ))\tilde{O}\big(s \cdot \mathrm{polylog}(d, \frac{1}{\epsilon})\big)O~(s⋅polylog(d,ϵ1​)) been established in [Zhang, 2018] for computationally efficient algorithms. In contrast, under high levels of label noise, the label complexity bounds achieved by computationally efficient algorithms are much worse: the best known result of [Awasthi et al., 2016] provides a computationally efficient algorithm with label complexity O~((sln⁡dϵ)2poly(1/(1−2η)))\tilde{O}\big((\frac{s \ln d}{\epsilon})^{2^{\mathrm{poly}(1/(1-2\eta))}} \big)O~((ϵslnd​)2poly(1/(1−2η))), which is label-efficient only when the noise rate η\etaη is a fixed constant. In this work, we substantially improve on it by designing a polynomial time algorithm for active learning of sss-sparse halfspaces, with a label complexity of O~(s(1−2η)4polylog(d,1ϵ))\tilde{O}\big(\frac{s}{(1-2\eta)^4} \mathrm{polylog} (d, \frac 1 \epsilon) \big)O~((1−2η)4s​polylog(d,ϵ1​)). This is the first efficient algorithm with label complexity polynomial in 11−2η\frac{1}{1-2\eta}1−2η1​ in this setting, which is label-efficient even for η\etaη arbitrarily close to 12\frac1221​. Our active learning algorithm and its theoretical guarantees also immediately translate to new state-of-the-art label and sample complexity results for full-dimensional active and passive halfspace learning under arbitrary bounded noise. The key insight of our algorithm and analysis is a new interpretation of online learning regret inequalities, which may be of independent interest.

View on arXiv
Comments on this paper