Chromatic Correlation Clustering (CCC) generalizes Correlation Clustering by assigning multiple categorical relationships (colors) to edges and imposing chromatic constraints on the clusters. Unlike traditional Correlation Clustering, which only deals with binary (+/-) relationships, CCC captures richer relational structures. Despite its importance, improving the approximation for CCC has been difficult due to the limitations of standard LP relaxations. We present a randomized 1.64-approximation algorithm to the CCC problem, significantly improving the previous factor of 2.15. Our approach extends the cluster LP framework to the chromatic setting by introducing a chromatic cluster LP relaxation and an rounding algorithm that utilizes both a cluster-based and a greedy pivot-based strategy. The analysis bypasses the integrality gap of 2 for the CCC version of standard LP and highlights the potential of the cluster LP framework to address other variants of clustering problems.

We study differentially private (DP) training with Muon, a matrix-valued optimizer that updates hidden-layer weights using momentum followed by Newton–Schulz orthogonalization. While DP-SGD is well understood, the interaction between per-example clipping, Gaussian noise, momentum, and nonlinear orthogonalization in Muon has not been systematically analyzed. We formulate DP-Muon, a private Muon procedure that clips per-example matrix gradients, adds Gaussian noise to the clipped lot average, and then applies momentum and Newton–Schulz orthogonalization as post-processing. We prove that DP-Muon inherits the privacy guarantee certified by the corresponding same-lot subsampled Gaussian accountant, with no additional privacy cost from Muon-specific post-processing. On the optimization side, we establish finite-horizon and vanishing stationarity guarantees under per-matrix clipping, with bounds that separate optimization error, clipping residual, privacy noise, and Newton–Schulz approximation error. We further show that the DP-induced bias in Muon arises not in the linear momentum buffer itself, but after the nonlinear Newton–Schulz map, where Gaussian noise induces a matrix-valued heat-smoothing bias. This motivates DP-MuonBC, a bias-corrected variant that removes the leading output-level bias term while preserving the same privacy guarantee. Experiments on E2E and DART show that Muon-style matrix updates improve private fine-tuning, and that DP-MuonBC further improves utility without increasing the privacy budget.

We provide the first study of the problem of finding differentially private (DP) second-order stationary points (SOSP) in stochastic (non-convex) minimax optimization. Existing literature either focuses only on first-order stationary points for minimax problems or on SOSP for classical stochastic minimization problems. This work provides, for the first time, a unified and detailed treatment of both empirical and population risks. Specifically, we propose a purely first-order method that combines a nested gradient descent--ascent scheme with SPIDER-style variance reduction and Gaussian perturbations to ensure privacy. A key technical device is a block-wise ($q$-period) analysis that controls the accumulation of stochastic variance and privacy noise without summing over the full iteration horizon, yielding a unified treatment of both empirical-risk and population formulations. Under standard smoothness, Hessian-Lipschitzness, and strong concavity assumptions, we establish high-probability guarantees for reaching an $(α,\sqrt{ρ_Φα})$-approximate second-order stationary point with $α= \mathcal{O}( (\frac{\sqrt{d}}{n\varepsilon})^{2/3})$ for empirical risk objectives and $\mathcal{O}(\frac{1}{n^{1/3}} + (\frac{\sqrt{d}}{n\varepsilon})^{1/2})$ for population objectives, matching the best known rates for private first-order stationarity.

We study the problem of releasing a synthetic graph that approximates the sizes of all cuts of an input graph under edge-level differential privacy. If one insists on purely additive error, the optimal worst-case error is (n^3/2). If one allows a small multiplicative slack, an information-theoretic exponential-time mechanism achieves nearly linear additive error, but the best known polynomial-time algorithms have substantially larger error. We give a polynomial-time (ε,δ)-differentially private algorithm which, for every n-vertex unweighted graph G, outputs a non-negative weighted synthetic graph G such that, with high probability, every cut S⊆ V(G) satisfies |w_G(S)-w_G(S)| ≤ γw_G(S)+O_ε,δ,γ(n^13/12+o(1)). This improves the previous polynomial-time worst-case bound O(n^5/4+o(1)) of Aamand et al. (ICML 2025) for mixed multiplicative/additive private cut approximation. The main technical ingredient is a new set of private spectral primitives for bounded-degree graphs, one of them gives spectral error O_δ((nd)^1/4/√(ε)) in estimating the graph Laplacian for graphs of maximum degree d, being the first to beat the standard min{2d,O_δ(√(n)/ε)} baseline in the high-degree regime. We further develop a primitive with a sharper error dependence on n and d for the downstream cut approximation. Combined with a new edge-sensitive terminal cut oracle with additive error O(n+(n^2M)^1/3) on graphs with M edges, this yields the final worst-case O(n^13/12+o(1)) private cut-release error.

In this paper, we address the challenge of differential privacy in the context of graph cuts, specifically focusing on the multiway cut and the minimum k-cut. We introduce edge-differentially private algorithms that achieve nearly optimal performance for these problems. Motivated by multiway cut, we propose the shifting mechanism, a general framework for private combinatorial optimization problems. This framework allows us to develop an efficient private algorithm with a multiplicative approximation ratio that matches the state-of-the-art non-private algorithm, improving over previous private algorithms that have provably worse multiplicative loss. We then provide a tight information-theoretic lower bound on the additive error, demonstrating that for constant k, our algorithm is optimal in terms of the privacy cost. The shifting mechanism also allows us to design private algorithm for the multicut and max-cut problems, with runtimes determined by the best non-private algorithms for these tasks. For the minimum k-cut problem we use a different approach, combining the exponential mechanism with bounds on the number of approximate k-cuts to get the first private algorithm with optimal additive error of O(klog n) (for a fixed privacy parameter). We also establish an information-theoretic lower bound that matches this additive error. Furthermore, we provide an efficient private algorithm even for non-constant k, including a polynomial-time 2-approximation with an additive error of Õ(k^1.5).

We study the problem of approximating all-pair distances in a weighted undirected graph with differential privacy, introduced by Sealfon [Sea16]. Given a publicly known undirected graph, we treat the weights of edges as sensitive information, and two graphs are neighbors if their edge weights differ in one edge by at most one. We obtain efficient algorithms with significantly improved bounds on a broad class of graphs which we refer to as recursively separable. In particular, for any n-vertex K_h-minor-free graph, our algorithm achieve an additive error of O(h(nW)^1/3 ), where W represents the maximum edge weight; For grid graphs, the same algorithmic scheme achieve additive error of O(n^1/4√(W)). Our approach can be seen as a generalization of the celebrated binary tree mechanism for range queries, as releasing range queries is equivalent to computing all-pair distances on a path graph. In essence, our approach is based on generalizing the binary tree mechanism to graphs that are recursively separable.

We study the problem of approximating all-pair distances in a weighted undirected graph with differential privacy, introduced by Sealfon [Sea16]. Given a publicly known undirected graph, we treat the weights of edges as sensitive information, and two graphs are neighbors if their edge weights differ in one edge by at most one. We obtain efficient algorithms with significantly improved bounds on a broad class of graphs which we refer to as *recursively separable*. In particular, for any $n$-vertex $K_h$-minor-free graph, our algorithm achieve an additive error of $ \widetilde{O}(h(nW)^{1/3} ) $, where $ W $ represents the maximum edge weight; For grid graphs, the same algorithmic scheme achieve additive error of $ \widetilde{O}(n^{1/4}\sqrt{W}) $. Our approach can be seen as a generalization of the celebrated binary tree mechanism for range queries, as releasing range queries is equivalent to computing all-pair distances on a path graph. In essence, our approach is based on generalizing the binary tree mechanism to graphs that are *recursively separable*.

We consider the Low Rank Approximation problem, where the input consists of a matrix $A \in \mathbb{R}^{n_R \times n_C}$ and an integer $k$, and the goal is to find a matrix $B$ of rank at most $k$ that minimizes $\| A - B \|_0$, which is the number of entries where $A$ and $B$ differ. For any constant $k$ and $\varepsilon > 0$, we present a polynomial time $(1 + \varepsilon)$-approximation time for this problem, which significantly improves the previous best $poly(k)$-approximation. Our algorithm is obtained by viewing the problem as a Constraint Satisfaction Problem (CSP) where each row and column becomes a variable that can have a value from $\mathbb{R}^k$. In this view, we have a constraint between each row and column, which results in a {\em dense} CSP, a well-studied topic in approximation algorithms. While most of previous algorithms focus on finite-size (or constant-size) domains and involve an exhaustive enumeration over the entire domain, we present a new framework that bypasses such an enumeration in $\mathbb{R}^k$. We also use tools from the rich literature of Low Rank Approximation in different objectives (e.g., $\ell_p$ with $p \in (0, \infty)$) or domains (e.g., finite fields/generalized Boolean). We believe that our techniques might be useful to study other real-valued CSPs and matrix optimization problems. On the hardness side, when $k$ is part of the input, we prove that Low Rank Approximation is NP-hard to approximate within a factor of $\Omega(\log n)$. This is the first superconstant NP-hardness of approximation for any $p \in [0, \infty]$ that does not rely on stronger conjectures (e.g., the Small Set Expansion Hypothesis).

While the standard unweighted Voronoi diagram in the plane has linear worst-case complexity, many of its natural generalizations do not. This paper considers two such previously studied generalizations, namely multiplicative and semi Voronoi diagrams. These diagrams both have quadratic worst-case complexity, though here we show that their expected complexity is linear for certain natural randomized inputs. Specifically, we argue that the expected complexity is linear for: (1) semi Voronoi diagrams when the visible direction is randomly sampled, and (2) for multiplicative diagrams when either weights are sampled from a constant-sized set, or the more challenging case when weights are arbitrary but locations are sampled from a square.

Private optimization is a topic of major interest in machine learning, with differentially private stochastic gradient descent (DP-SGD) playing a key role in both theory and practice. Furthermore, DP-SGD is known to be a powerful tool in contexts beyond privacy, including robustness, machine unlearning, etc. Existing analyses of DP-SGD either make relatively strong assumptions (e.g., Lipschitz continuity of the loss function, or even convexity) or prove only first-order convergence (and thus might end at a saddle point in the non-convex setting). At the same time, there has been progress in proving second-order convergence of the non-private version of “noisy SGD”, as well as progress in designing algorithms that are more complex than DP-SGD and do guarantee second-order convergence. We revisit DP-SGD and show that “noise is all you need”: the noise necessary for privacy already implies second-order convergence under the standard smoothness assumptions, even for non-Lipschitz loss functions. Hence, we get second-order convergence essentially for free: DP-SGD, the workhorse of modern private optimization, under minimal assumptions can be used to find a second-order stationary point.

When designing clustering algorithms, the choice of initial centers is crucial for the quality of the learned clusters. In this paper, we develop a new initialization scheme, called HST initialization, for the k-median problem in the general metric space (e.g., discrete space induced by graphs), based on the construction of metric embedding tree structure of the data. From the tree, we propose a novel and efficient search algorithm, for good initial centers that can be used subsequently for the local search algorithm. Our proposed HST initialization can produce initial centers achieving lower errors than those from another popular initialization method, k-median++, with comparable efficiency. The HST initialization can also be extended to the setting of differential privacy (DP) to generate private initial centers. We show that the error from applying DP local search followed by our private HST initialization improves previous results on the approximation error, and approaches the lower bound within a small factor. Experiments justify the theory and demonstrate the effectiveness of our proposed method. Our approach can also be extended to the k-means problem.

The densest subgraph of a large graph usually refers to some subgraph with the highest average degree, which has been extended to the family of p-means dense subgraph objectives by . The p-mean densest subgraph problem seeks a subgraph with the highest average p-th-power degree, whereas the standard densest subgraph problem seeks a subgraph with a simple highest average degree. It was shown that the standard peeling algorithm can perform arbitrarily poorly on generalized objective when p>1 but uncertain when 0

k -means clustering is an important problem in machine learning and statistics. The k -means ++ initialization algorithm has driven new acceleration strategies and theoretical analysis for solving the k -means clustering problem. The state-of-the-art variant, called LocalSearch ++, adds extra local search steps upon k -means ++ to achieve constant approximation error in expectation. In this paper, we propose a new variant named LSDS ++, which improves the sampling efficiency of LocalSearch ++ via a strategy called dual sampling . By defining a new capture graph based on the concept of coreset, we show that the proposed LSDS ++ is able to achieve the same expected constant error with reduced complexity. Experiments are conducted to justify the benefit of LSDS ++ in practice.

In this article, we study a wide range of variants for computing the (discrete and continuous) Fréchet distance between uncertain curves. An uncertain curve is a sequence of uncertainty regions, where each region is a disk, a line segment, or a set of points. A realisation of a curve is a polyline connecting one point from each region. Given an uncertain curve and a second (certain or uncertain) curve, we seek to compute the lower and upper bound Fréchet distance, which are the minimum and maximum Fréchet distance for any realisations of the curves. We prove that both problems are NP-hard for the Fréchet distance in several uncertainty models, and that the upper bound problem remains hard for the discrete Fréchet distance. In contrast, the lower bound (discrete [ 5 ] and continuous) Fréchet distance can be computed in polynomial time in some models. Furthermore, we show that computing the expected (discrete and continuous) Fréchet distance is #P-hard in some models. On the positive side, we present an FPTAS in constant dimension for the lower bound problem when Δ/δ is polynomially bounded, where δ is the Fréchet distance and Δ bounds the diameter of the regions. We also show a near-linear-time 3-approximation for the decision problem on roughly δ-separated convex regions. Finally, we study the setting with Sakoe–Chiba time bands, where we restrict the alignment between the curves, and give polynomial-time algorithms for the upper bound and expected discrete and continuous Fréchet distance for uncertainty modelled as point sets.

Differential private stochastic gradient descent (DP-SGD) with gradient clipping (DP-SGD-GC) is an effective optimization algorithm that can train machine learning models with a privacy guarantee. Despite the popularity of DP-SGD-GC, its convergence in unbounded domain without the Lipschitz continuous assumption is less-understood; existing analysis of DP-SGD-GC either impose additional assumptions or end up with an utility bound that involves an non-vanishing bias term. In this work, for smooth and unconstrained problems, we improve the current analysis and show that DP-SGD-GC can achieve a vanishing utility bound without any bias term. Furthermore, when the noise generated from subsampled gradients is light-tailed, we prove that DP-SGD-GC can achieve nearly the same utility bound as DP-SGD applies to the Lipschitz continuous objectives. As a by-product, we propose a new clipping technique, called value clipping, to mitigate the computational overhead caused by the classic gradient clipping. Experiments on standard benchmark datasets are conducted to support our analysis.

Releasing all pairwise shortest path (APSP) distances between vertices on general graphs under weight Differential Privacy (DP) is known as a challenging task. In previous work, to achieve DP with some fixed budget, with high probability the maximal absolute error among all published pairwise distances is roughly O(n) where n is the number of nodes. It was shown that this error could be reduced for some special graphs, which, however, is hard for general graphs. Therefore, whether the approximation error can be reduced to sublinear is posted as an interesting open problem. In this paper, we break the linear barrier on the distance approximation error of previous result, by proposing an algorithm that releases a constructed synthetic graph privately. Computing all pairwise distances on the constructed graph only introduces O(n^{1/2}) error in answering all pairwise shortest path distances for fixed privacy parameter. Our method is based on a novel graph diameter (link length) augmentation via constructing ``shortcuts'' for the paths. By adding a set of shortcut edges to the original graph, we show that any node pair has a shortest path with link length O(n^{1/2}). Then by adding noises with some positive mean to the edge weights, the new graph is differentially private and can be published to answer all pairwise shortest path distances with O(n^{1/2}) approximation error using standard APSP computation. Numerical examples are also provided. Additionally, we also consider the graph with small feedback vertex set number. A feedback vertex set (FVS) of a graph is a set of vertices whose removal leaves a graph without cycles, and the feedback vertex set number of a graph, k, is the size of a smallest feedback vertex set. We propose a DP algorithm with error rate O(k), which improves the error of general graphs provided k=o(n^{1/2}).

We study the facility location problem under the constraints imposed by local differential privacy (LDP). Recently, Gupta et al. (2010) and Esencayi et al. (2019) proposed lower and upper bounds for the problem on the central differential privacy (DP) model where a trusted curator first collects all data and processes it. In this paper, we focus on the LDP model, where we protect a client's participation in the facility location instance. Under the HST metric, we show that there is a non-interactive epsilon-LDP algorithm achieving O(n(1/4)/epsilon(2))-approximation ratio, where n is the size of the metric. On the negative side, we show a lower bound of Omega(n(1/4)/root epsilon) on the approximation ratio for any non-interactive epsilon-LDP algorithm. Thus, our results are tight up to a polynomial factor of epsilon. Moreover, unlike previous results, our results generalize to non-uniform facility costs.

Given $x\in(\mathbb{R}_{\geqslant 0})(_{2}^{[n]})$ recording pairwise distances, the Metric Violation Distance problem asks to compute the $\ell_{0}$ distance between x and the metric cone; i.e., modify the minimum number of entries of x to make it a metric. Due to its large number of applications in various data analysis and optimization tasks, this problem has been actively studied recently. We present an $O(\log n)$-approximation algorithm for METRIC VIOLATION Distance, exponentially improving the previous best approximation ratio of $O(OPT^{1/3})$ of Fan, Raichel, and Van Buskirk [SODA, 2018]. Furthermore, a major strength of our algorithm is its simplicity and running time. We also study the related problem of Ultrametric Violation Distance, where the goal is to compute the $\ell_{0}$ distance to the cone of ultrametrics, and achieve a constant factor approximation algorithm. The ULTRAMETRIC VIOLATION DISTANCE problem can be regarded as an extension of the problem of fitting ultrametrics studied by Ailon and Charikar [SIAM J. Computing, 2011] and by Cohen-Addad, Das, Kipouridis, Parotsidis, and Thorup [FOCS, 2021] from $\ell_{1}$ norm to $\ell_{0}$ norm. We show that this problem can be favorably interpreted as an instance of CORRELATION CLUSTERING with an additional hierarchical structure, which we solve using a new $O(1)$-approximation algorithm for correlation clustering that has the structural property that it outputs a refinement of the optimum clusters. An algorithm satisfying such a property can be considered of independent interest. We also provide an $O(\log n\log\log n)$ approximation algorithm for weighted instances. Finally, we investigate the complementary version of these problems where one aims at choosing a maximum number of entries of x forming an (ultra-)metric. In stark contrast with the minimization versions, we prove that these maximization versions are hard to approximate within any constant factor assuming the Unique Games Conjecture.

Data about individuals may contain private and sensitive information. The differential privacy (DP) was proposed to address the problem of protecting the privacy of each individual while keeping useful information about a population. Sealfon [1] introduced a private graph model in which the graph topology is assumed to be public while the weight information is assumed to be private. That model can express hidden congestion patterns in a known transportation system. In this paper, we revisit the problem of privately releasing approximate distances between all pairs of vertices in [1]. Our goal is to minimize the additive error, namely the difference between the released distance and actual distance under private setting. We propose improved solutions to that problem for several cases.For the problem of privately releasing all-pairs distances, we show that for tree with depth h, we can release all-pairs distances with additive error O(log 1.5 h • log 1.5 V) for fixed privacy parameter where V the number of vertices in the tree, which improves the previous error bound O(log 2.5 V), since the size of h can be as small as O(log V). Our result implies that a log V factor is saved, and the additive error in tree can be smaller than the error on array/path. Additionally, for the grid graph with arbitrary edge weights, we also propose a method to release all-pairs distances with additive error $\tilde O\left( {{V^{3/4}}} \right)$ for fixed privacy parameters. On the application side, many cities like Manhattan are composed of horizontal streets and vertical avenues, which can be modeled as a grid graph.