We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.CC

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Computational Complexity

Title: Hardness of Maximum Likelihood Learning of DPPs

Abstract: Determinantal Point Processes (DPPs) are a widely used probabilistic model for negatively correlated sets. DPPs have been successfully employed in Machine Learning applications to select a diverse, yet representative subset of data. In seminal work on DPPs in Machine Learning, Kulesza conjectured in his PhD Thesis (2011) that the problem of finding a maximum likelihood DPP model for a given data set is NP-complete.
In this work we prove Kulesza's conjecture. In fact, we prove the following stronger hardness of approximation result: even computing a $\left(1-O(\frac{1}{\log^9{N}})\right)$-approximation to the maximum log-likelihood of a DPP on a ground set of $N$ elements is NP-complete. At the same time, we also obtain the first polynomial-time algorithm that achieves a nontrivial worst-case approximation to the optimal log-likelihood: the approximation factor is $\frac{1}{(1+o(1))\log{m}}$ unconditionally (for data sets that consist of $m$ subsets), and can be improved to $1-\frac{1+o(1)}{\log N}$ if all $N$ elements appear in a $O(1/N)$-fraction of the subsets.
In terms of techniques, we reduce approximating the maximum log-likelihood of DPPs on a data set to solving a gap instance of a "vector coloring" problem on a hypergraph. Such a hypergraph is built on a bounded-degree graph construction of Bogdanov, Obata and Trevisan (FOCS 2002), and is further enhanced by the strong expanders of Alon and Capalbo (FOCS 2007) to serve our purposes.
Subjects: Computational Complexity (cs.CC); Data Structures and Algorithms (cs.DS); Machine Learning (cs.LG)
Cite as: arXiv:2205.12377 [cs.CC]
  (or arXiv:2205.12377v2 [cs.CC] for this version)

Submission history

From: Ning Xie [view email]
[v1] Tue, 24 May 2022 21:46:23 GMT (54kb)
[v2] Thu, 26 May 2022 02:12:36 GMT (54kb)

Link back to: arXiv, form interface, contact.