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. 1304.2079
121
41
v1v2v3 (latest)

Learning Coverage Functions and Private Release of Marginals

8 April 2013
Vitaly Feldman
Pravesh Kothari
ArXiv (abs)PDFHTML
Abstract

We study the problem of approximating and learning coverage functions. A function c:2[n]→R+c: 2^{[n]} \rightarrow \mathbf{R}^{+}c:2[n]→R+ is a coverage function, if there exists a universe UUU with non-negative weights w(u)w(u)w(u) for each u∈Uu \in Uu∈U and subsets A1,A2,…,AnA_1, A_2, \ldots, A_nA1​,A2​,…,An​ of UUU such that c(S)=∑u∈∪i∈SAiw(u)c(S) = \sum_{u \in \cup_{i \in S} A_i} w(u)c(S)=∑u∈∪i∈S​Ai​​w(u). Alternatively, coverage functions can be described as non-negative linear combinations of monotone disjunctions. They are a natural subclass of submodular functions and arise in a number of applications. We give an algorithm that for any γ,δ>0\gamma,\delta>0γ,δ>0, given random and uniform examples of an unknown coverage function ccc, finds a function hhh that approximates ccc within factor 1+γ1+\gamma1+γ on all but δ\deltaδ-fraction of the points in time poly(n,1/γ,1/δ)poly(n,1/\gamma,1/\delta)poly(n,1/γ,1/δ). This is the first fully-polynomial algorithm for learning an interesting class of functions in the demanding PMAC model of Balcan and Harvey (2011). Our algorithms are based on several new structural properties of coverage functions. Using the results in (Feldman and Kothari, 2014), we also show that coverage functions are learnable agnostically with excess ℓ1\ell_1ℓ1​-error ϵ\epsilonϵ over all product and symmetric distributions in time nlog⁡(1/ϵ)n^{\log(1/\epsilon)}nlog(1/ϵ). In contrast, we show that, without assumptions on the distribution, learning coverage functions is at least as hard as learning polynomial-size disjoint DNF formulas, a class of functions for which the best known algorithm runs in time 2O~(n1/3)2^{\tilde{O}(n^{1/3})}2O~(n1/3) (Klivans and Servedio, 2004). As an application of our learning results, we give simple differentially-private algorithms for releasing monotone conjunction counting queries with low average error. In particular, for any k≤nk \leq nk≤n, we obtain private release of kkk-way marginals with average error αˉ\bar{\alpha}αˉ in time nO(log⁡(1/αˉ))n^{O(\log(1/\bar{\alpha}))}nO(log(1/αˉ)).

View on arXiv
Comments on this paper