Searcharxiv⌕ Search

arXiv subjects

Zachary Friggstad

Publications and source records attributed to Zachary Friggstad.

At least 19 recordsLinked to original sources

Minimum Sum-of-Radii Clustering with Strong and Weak Oracles

In the minimum sum of radii (MSR) problem, the input is a set of $n$ points in a metric space and a parameter $k$, and the goal is to cover the points by at most $k$ balls, each centered at an input point, while minimizing the sum of their radii. We study MSR in the weak-strong oracle model of Bateni, Dharangutte, Jayaram, and Wang (COLT'24), where the metric is unknown and is accessed through two oracles. The answers of the weak oracle are given to us for free: for every pair of points, we receive a distance value that, with probability $δ$, is arbitrarily corrupted. Exact distances are available only through costly strong queries, where a strong query on a point reveals its exact distances to all previously strongly queried points, and the goal is to minimize the number of strong queries. For every constant $δ< 1/2$, we give the first polynomial-time algorithm that makes $O(k \log n \log\log n)$ strong point queries and, with high probability, returns an $O(1)$-approximation for MSR; the previously best algorithm required running time and a number of strong queries exponential in $k$. This is within an $O(\log\log n)$ factor of a lower bound that we prove: any constant-factor approximation that succeeds with high probability on every instance must make $Ω(k \log n)$ strong queries in expectation on some instance. The same $Ω(k \log n)$ lower bound holds for $k$-center, where it matches the known upper bound up to an $O(\log\log n)$ factor. An optimal MSR cluster can be small both in size and in radius, so a uniform sample of strongly queried points may miss it entirely. Instead, our algorithm uses a packing LP over the balls certified so far to decide which points to query next. The analysis holds against an adversary who knows which pairs are corrupted.

cs.DS↗

A PTAS for Euclidean Capacitated Vehicle Routing

In the Capacitated Vehicle Routing Problem (CVRP), introduced by Dantzig and Ramser in 1959, we are given an $n$-node graph $G=(V,E)$ with metric edges costs $c(e)$, a depot $r\in V$, and an integer vehicle capacity $Q$. The goal is to find a minimum-cost collection of tours, each containing the depot and visiting at most other $Q$ nodes, such that the tours cover all the nodes. In this paper we focus on the Euclidean CVRP problem, where the nodes represent points on $\mathbb{R}^2$, and the edge costs are the respective Euclidean distances. Euclidean CVRP generalizes the classical NP-hard Euclidean TSP problem (the special case with $Q=n-1$), that admits a PTAS [Arora FOCS'96/Mitchell SODA'96]. A long-standing open problem has been generalizing the latter PTAS to Euclidean CVRP. A polynomial-time $(2+ε)$-approximation for this problem follows from the PTAS for Euclidean TSP and the tour-partitioning technique for CVRP [Haimovich, Kan Math. OR'85]. This was the best known until a recent improvement of the approximation factor to slightly below 2 [Friggstad, Grandoni, Mousavi SODA'26]. There have been a series of papers presenting improved QPTASs by Das and Mathieu [SODA'10] (time $n^{\log^{O(1/ε)}n}$), Jayaprakash and Salavatipour [SODA'22] (time $n^{O_ε(\log^6 n)}$), and recently Sitters [ArXiV'25] (time $n^{O_ε(\log\log n)}$). In this paper, we solve the mentioned long-standing open problem by presenting a PTAS for Euclidean CVRP.

cs.DS↗

Multiple Swap Local Search Approximations for Uniform-Cost Capacitated Facility Location with Extensions to Outliers

We study \textsc{Capacitated Facility Location}, a fundamental clustering problem, in the setting of uniform opening costs and non-uniform capacities. We give a polynomial-time $(3+ε)$-approximation for this problem. Our algorithm is a multiple-swap generalization of the local search algorithm by Dabas, Garg, and Gupta (2024), who obtained a $3.732$-approximation. The main technical contribution is the analysis of this natural multiple-swap local-search method: variants of this approach have been analyzed previously, but allowing larger swaps requires new grouping and counting arguments. We also show that a multiple-swap local search algorithm provides a $(3+ε)$-approximation for \textsc{Capacitated Facility Location with Outliers}, again in the same setting of uniform opening costs and non-uniform capacities. This is an improvement over the $6.372$-approximation, again by Dabas, Garg, and Gupta. Finally, we show that the factor $3$ is tight for our local-search framework, and that the same multiple-swap idea does not improve the known locality gap for the non-uniform opening-cost setting.

cs.DS↗

Facility Location with Customer Incompatibilities

We study metric facility location with customer incompatibilities, where the clients assigned to each facility must satisfy prescribed independence constraints. We first consider the setting of facility-opening costs and where there is a single matroid $\mathcal M$ over the clients and the set of clients assigned to an open facility must be an independent set in $\mathcal M$. We develop a polynomial-time \((1+\sqrt{2}+ε)\)-approximation algorithm for this setting. We further obtain a polynomial-time \((15+ε)\)-approximation algorithm for the case of non-uniform facility-opening costs when $\mathcal M$ is a partition matroid that only permits $b_g$ clients to be assigned to a facility for each group $g$ in the partition. Both settings are substantial generalizations of Capacitated Facility Location with uniform capacities. Our algorithms are based on local search using add, delete, and single-swap operations. As with CFL, bounding the connection cost is reasonably straightforward using supermodularity of the minimum connection-cost function. However, bounding the facility opening costs requires novel arguments in both cases. We use a ``compression'' argument based on matroid exchange principles to identify a subset of the locally opened facilities whose size is no larger than that of an optimal facility set. The opening costs of facilities outside this subset can be bounded using supermodularity, while bounding the costs of the remaining facilities requires different arguments. It is natural to consider a variant where each possible facility $i$ has its own matroid $\mathcal M_i$ defining the subsets of clients it can feasibly serve. We also show problem is at least as hard to approximate as Set Cover even if all $\mathcal M_i$ are partition matroids that have the same partition of clients, just different group bounds $b_g$.

cs.DS↗

$k$-Center under Weak and Strong Distance Oracles with an Application to Minimum Sum of Radii Clustering

In the weak-strong distance oracle model of Bateni, Dharangutte, Jayaram and Wang (COLT 2024), a metric $(X,d)$ on $n$ points is accessed through two oracles: a cheap weak oracle that returns each queried distance correctly with probability $1-δ$ and permits arbitrary values on corrupted pairs, chosen by an adversary who knows the entire metric and all corrupted pairs, and an expensive strong oracle that reveals a point exactly. The goal is to cluster well while making as few strong queries as possible. Our main result is a $(6+\varepsilon)$-approximation for $k$-center using $O(k\log n\log(\varepsilon^{-1}\log n))$ strong point queries, for any fixed $\varepsilon>0$. This improves both the $(14+\varepsilon)$ approximation factor and the $O(k\log^2 n\log(\varepsilon^{-1}\log n))$ query bound of Bateni et al. It also improves the query complexity of the recent $(6+\varepsilon)$-approximation of Pradhan, Bhattacharya and Jaiswal (AISTATS 2026), in the variant of the model they study. Our algorithm outputs an explicitly known sketch of an otherwise unknown metric: a concrete metric on at most $k$ points that a downstream algorithm can compute with directly, with no further queries and no noise. As an application, we give a simple FPT algorithm for Minimum Sum of Radii (MSR) that returns a constant-factor solution using $2^{O(k\log k)}\cdot O(\log n\log\log n)$ strong queries. We complement the upper bound with an $Ω(k\log(n/k))$ lower bound for $k$-center with assignment that holds for row-local algorithms. These algorithms assign two unqueried points to the same center whenever their weak distances to the strong-queried points are exactly the same. Our algorithm and those of Bateni et al. and Pradhan et al. are all row-local, so our query bound is optimal for this class up to the $O(\log\log n)$ factor.

cs.DS↗

Approximating Traveling Salesman Problems Using a Bridge Lemma

We give improved approximations for two metric Traveling Salesman Problem (TSP) variants. In Ordered TSP (OTSP) we are given a linear ordering on a subset of nodes $o_1, \ldots, o_k$. The TSP solution must have that $o_{i+1}$ is visited at some point after $o_i$ for each $1 \leq i < k$. This is the special case of Precedence-Constrained TSP ($PTSP$) in which the precedence constraints are given by a single chain on a subset of nodes. In $k$-Person TSP Path (k-TSPP), we are given pairs of nodes $(s_1,t_1), \ldots, (s_k,t_k)$. The goal is to find an $s_i$-$t_i$ path with minimum total cost such that every node is visited by at least one path. We obtain a $3/2 + e^{-1} < 1.878$ approximation for OTSP, the first improvement over a trivial $α+1$ approximation where $α$ is the current best TSP approximation. We also obtain a $1 + 2 \cdot e^{-1/2} < 2.214$ approximation for k-TSPP, the first improvement over a trivial $3$-approximation. These algorithms both use an adaptation of the Bridge Lemma that was initially used to obtain improved Steiner Tree approximations [Byrka et al., 2013]. Roughly speaking, our variant states that the cost of a cheapest forest rooted at a given set of terminal nodes will decrease by a substantial amount if we randomly sample a set of non-terminal nodes to also become terminals such provided each non-terminal has a constant probability of being sampled. We believe this view of the Bridge Lemma will find further use for improved vehicle routing approximations beyond this paper.

cs.DS↗

Budget-constrained Active Learning to Effectively De-censor Survival Data

Standard supervised learners attempt to learn a model from a labeled dataset. Given a small set of labeled instances, and a pool of unlabeled instances, a budgeted learner can use its given budget to pay to acquire the labels of some unlabeled instances, which it can then use to produce a model. Here, we explore budgeted learning in the context of survival datasets, which include (right) censored instances, where we know only a lower bound on an instance's time-to-event. Here, that learner can pay to (partially) label a censored instance -- e.g., to acquire the actual time for an instance [perhaps go from (3 yr, censored) to (7.2 yr, uncensored)], or other variants [e.g., learn about one more year, so go from (3 yr, censored) to either (4 yr, censored) or perhaps (3.2 yr, uncensored)]. This serves as a model of real world data collection, where follow-up with censored patients does not always lead to uncensoring, and how much information is given to the learner model during data collection is a function of the budget and the nature of the data itself. We provide both experimental and theoretical results for how to apply state-of-the-art budgeted learning algorithms to survival data and the respective limitations that exist in doing so. Our approach provides bounds and time complexity asymptotically equivalent to the standard active learning method BatchBALD. Moreover, empirical analysis on several survival tasks show that our model performs better than other potential approaches on several benchmarks.

cs.LG↗

Approximating Multiple-Depot Capacitated Vehicle Routing via LP Rounding

In Capacitated Vehicle Routing with Multiple Depots (CVRP-MD) we are given a set of client locations $C$ and a set of depots $R$ located in a metric space with costs $c(i,j)$ between $u,v \in C \cup R$. Additionally, we are given a capacity bound $k$. The goal is to find a collection of tours of minimum total cost such that each tour starts and ends at some depot $r \in R$ and includes at most $k$ clients and such that each client lies on at least one tour. Our main result is a $3.9365$-approximation based on rounding a new LP relaxation for CVRP-MD.

cs.DS↗

A Polynomial-Time Approximation for Pairwise Fair $k$-Median Clustering

In this work, we study pairwise fair clustering with $\ell \ge 2$ groups, where for every cluster $C$ and every group $i \in [\ell]$, the number of points in $C$ from group $i$ must be at most $t$ times the number of points in $C$ from any other group $j \in [\ell]$, for a given integer $t$. To the best of our knowledge, only bi-criteria approximation and exponential-time algorithms follow for this problem from the prior work on fair clustering problems when $\ell > 2$. In our work, focusing on the $\ell > 2$ case, we design the first polynomial-time $O(k^2\cdot \ell \cdot t)$-approximation for this problem with $k$-median cost that does not violate the fairness constraints. We complement our algorithmic result by providing hardness of approximation results, which show that our problem even when $\ell=2$ is almost as hard as the popular uniform capacitated $k$-median, for which no polynomial-time algorithm with an approximation factor of $o(\log k)$ is known.

cs.DS↗

Approximation Algorithms for Clustering with Minimum Sum of Radii, Diameters, and Squared Radii

In this paper, we present an improved approximation algorithm for three related problems. In the Minimum Sum of Radii clustering problem (MSR), we aim to select $k$ balls in a metric space to cover all points while minimizing the sum of the radii. In the Minimum Sum of Diameters clustering problem (MSD), we are to pick $k$ clusters to cover all the points such that sum of diameters of all the clusters is minimized. At last, in the Minimum Sum of Squared Radii problem (MSSR), the goal is to choose $k$ balls, similar to MSR. However in MSSR, the goal is to minimize the sum of squares of radii of the balls. We present a 3.389-approximation for MSR and a 6.546-approximation for MSD, improving over respective 3.504 and 7.008 developed by Charikar and Panigrahy (2001). In particular, our guarantee for MSD is better than twice our guarantee for MSR. In the case of MSSR, the best known approximation guarantee is $4\cdot(540)^{2}$ based on the work of Bhowmick, Inamdar, and Varadarajan in their general analysis of the $t$-Metric Multicover Problem. At last with our analysis, we get a 11.078-approximation algorithm for Minimum Sum of Squared Radii.

cs.DS↗

A QPTAS for Facility Location on Unit Disk graphs

We study the classic \textsc{(Uncapacitated) Facility Location} problem on Unit Disk Graphs (UDGs). For a given point set $P$ in the plane, the unit disk graph UDG(P) on $P$ has vertex set $P$ and an edge between two distinct points $p, q \in P$ if and only if their Euclidean distance $|pq|$ is at most 1. The weight of the edge $pq$ is equal to their distance $|pq|$. An instance of \fl on UDG(P) consists of a set $C\subseteq P$ of clients and a set $F\subseteq P$ of facilities, each having an opening cost $f_i$. The goal is to pick a subset $F'\subseteq F$ to open while minimizing $\sum_{i\in F'} f_i + \sum_{v\in C} d(v,F')$, where $d(v,F')$ is the distance of $v$ to nearest facility in $F'$ through UDG(P). In this paper, we present the first Quasi-Polynomial Time Approximation Schemes (QPTAS) for the problem. While approximation schemes are well-established for facility location problems on sparse geometric graphs (such as planar graphs), there is a lack of such results for dense graphs. Specifically, prior to this study, to the best of our knowledge, there was no approximation scheme for any facility location problem on UDGs in the general setting.

cs.DS↗

Exact Algorithms and Lower Bounds for Stable Instances of Euclidean k-Means

We investigate the complexity of solving stable or perturbation-resilient instances of $k$-Means and $k$-Median clustering in fixed dimension Euclidean metrics (more generally doubling metrics). The notion of stable (perturbation resilient) instances was introduced by Bilu and Linial [2010] and Awasthi et al. [2012]. In our context we say a $k$-Means instance is $α$-stable if there is a unique OPT which remains optimum if distances are (non-uniformly) stretched by a factor of at most $α$. Stable clustering instances have been studied to explain why heuristics such as Lloyd's algorithm perform well in practice. In this work we show that for any fixed $ε>0$, $(1+ε)$-stable instances of $k$-Means in doubling metrics can be solved in polynomial time. More precisely we show a natural multiswap local search algorithm finds OPT for $(1+ε)$-stable instances of $k$-Means and $k$-Median in a polynomial number of iterations. We complement this result by showing that under a new PCP theorem, this is essentially tight: that when the dimension d is part of the input, there is a fixed $ε_0>0$ s.t. there is not even a PTAS for $(1+ε_0)$-stable $k$-Means in $R^d$ unless NP=RP. To do this, we consider a robust property of CSPs; call an instance stable if there is a unique optimum solution $x^*$ and for any other solution $x'$, the number of unsatisfied clauses is proportional to the Hamming distance between $x^*$ and $x'$. Dinur et al. have already shown stable QSAT is hard to approximate for some constant Q, our hypothesis is simply that stable QSAT with bounded variable occurrence is also hard. Given this hypothesis we consider "stability-preserving" reductions to prove our hardness for stable k-Means. Such reductions seem to be more fragile than standard L-reductions and may be of further use to demonstrate other stable optimization problems are hard.

cs.DS↗

An $O(\log k)$-Approximation for Directed Steiner Tree in Planar Graphs

We present an $O(\log k)$-approximation for both the edge-weighted and node-weighted versions of \DST in planar graphs where $k$ is the number of terminals. We extend our approach to \MDST (in general graphs \MDST and \DST are easily seen to be equivalent but in planar graphs this is not the case necessarily) in which we get an $O(R+\log k)$-approximation for planar graphs for where $R$ is the number of roots.

cs.DS↗

On Narrative Information and the Distillation of Stories

The act of telling stories is a fundamental part of what it means to be human. This work introduces the concept of narrative information, which we define to be the overlap in information space between a story and the items that compose the story. Using contrastive learning methods, we show how modern artificial neural networks can be leveraged to distill stories and extract a representation of the narrative information. We then demonstrate how evolutionary algorithms can leverage this to extract a set of narrative templates and how these templates -- in tandem with a novel curve-fitting algorithm we introduce -- can reorder music albums to automatically induce stories in them. In the process of doing so, we give strong statistical evidence that these narrative information templates are present in existing albums. While we experiment only with music albums here, the premises of our work extend to any form of (largely) independent media.

cs.CL↗

A Constant-Factor Approximation for Quasi-bipartite Directed Steiner Tree on Minor-Free Graphs

We give the first constant-factor approximation algorithm for quasi-bipartite instances of Directed Steiner Tree on graphs that exclude fixed minors. In particular, for $K_r$-minor-free graphs our approximation guarantee is $O(r\cdot\sqrt{\log r})$ and, further, for planar graphs our approximation guarantee is 20. Our algorithm uses the primal-dual scheme. We employ a more involved method of determining when to buy an edge while raising dual variables since, as we show, the natural primal-dual scheme fails to raise enough dual value to pay for the purchased solution. As a consequence, we also demonstrate integrality gap upper bounds on the standard cut-based linear programming relaxation for the Directed Steiner Tree instances we consider.

cs.DS↗

Parameterized Approximation Algorithms for $k$-Center Clustering and Variants

$k$-center is one of the most popular clustering models. While it admits a simple 2-approximation in polynomial time in general metrics, the Euclidean version is NP-hard to approximate within a factor of 1.93, even in the plane, if one insists the dependence on $k$ in the running time be polynomial. Without this restriction, a classic algorithm yields a $2^{O((k\log k)/ε)}dn$-time $(1+ε)$-approximation for Euclidean $k$-center, where $d$ is the dimension. We give a faster algorithm for small dimensions: roughly speaking an $O^*(2^{O((1/ε)^{O(d)} \cdot k^{1-1/d} \cdot \log k)})$-time $(1+ε)$-approximation. In particular, the running time is roughly $O^*(2^{O((1/ε)^{O(1)}\sqrt{k}\log k)})$ in the plane. We complement our algorithmic result with a matching hardness lower bound. We also consider a well-studied generalization of $k$-center, called Non-uniform $k$-center (NUkC), where we allow different radii clusters. NUkC is NP-hard to approximate within any factor, even in the Euclidean case. We design a $2^{O(k\log k)}n^2$ time $3$-approximation for NUkC in general metrics, and a $2^{O((k\log k)/ε)}dn$ time $(1+ε)$-approximation for Euclidean NUkC. The latter time bound matches the bound for $k$-center.

cs.DS↗

Improved Approximations for CVRP with Unsplittable Demands

In this paper, we present improved approximation algorithms for the (unsplittable) Capacitated Vehicle Routing Problem (CVRP) in general metrics. In CVRP, introduced by Dantzig and Ramser (1959), we are given a set of points (clients) $V$ together with a depot $r$ in a metric space, with each $v\in V$ having a demand $d_v>0$, and a vehicle of bounded capacity $Q$. The goal is to find a minimum cost collection of tours for the vehicle, each starting and ending at the depot, such that each client is visited at least once and the total demands of the clients in each tour is at most $Q$. In the unsplittable variant we study, the demand of a node must be served entirely by one tour. We present two approximation algorithms for unsplittable CVRP: a combinatorial $(α+1.75)$-approximation, where $α$ is the approximation factor for the Traveling Salesman Problem, and an approximation algorithm based on LP rounding with approximation guarantee $α+\ln(2) + δ\approx 3.194 + δ$ in $n^{O(1/δ)}$ time. Both approximations can further be improved by a small amount when combined with recent work by Blauth, Traub, and Vygen (2021), who obtained an $(α+ 2\cdot (1 -ε))$-approximation for unsplittable CVRP for some constant $ε$ depending on $α$ ($ε> 1/3000$ for $α= 1.5$).

cs.DS↗