arXiv subjects

Michael Dinitz

Publications and source records attributed to Michael Dinitz.

At least 19 recordsLinked to original sources

arXiv2026

Tree Search With Predictions

``Algorithms with predictions'', or ``learning-augmented algorithms'', has proved to be an extremely useful paradigm for combining machine learning with traditional algorithms. One of the textbook settings for this is searching a sorted array. Without a prediction, classical binary search takes $O(\log n)$ queries, while with a prediction we can use ``doubling binary search'' to find the target key using $O(\log \eta)$ queries, where $\eta$ is the error of the prediction measured as the absolute value of the difference between the true location and the predicted location. Since an array is just a path graph, in this paper we ask whether similar bounds can be achieved for search on even slightly more general graphs: trees. We show first that the high-level answer is ``no'': there is no search algorithm that uses $O(\log \eta)$ queries, where $\eta$ is now the graph distance between the predicted location and the true location. However, as our main result, we show that such bounds can be achieved on trees which are ``path-like'' in that they have low \emph{pathwidth}. In particular, we prove that there is a search algorithm which uses at most $O(k \log \eta)$ queries, where $k$ is the pathwidth of the tree. We also prove a lower bound showing that our algorithm has existentially optimal query complexity. Finally, we show experimentally, on real-life inputs, that our algorithm has query complexity which is notably better than the simple non-prediction-based algorithm.

cs.DS↗

arXiv2026

Ski Rental with Distributional Predictions of Unknown Quality

We revisit the central online problem of ski rental in the "algorithms with predictions" framework from the point of view of distributional predictions. Ski rental was one of the first problems to be studied with predictions, where a natural prediction is simply the number of ski days. But it is both more natural and potentially more powerful to think of a prediction as a distribution p-hat over the ski days. If the true number of ski days is drawn from some true (but unknown) distribution p, then we show as our main result that there is an algorithm with expected cost at most OPT + O(min(max({eta}, 1) * sqrt(b), b log b)), where OPT is the expected cost of the optimal policy for the true distribution p, b is the cost of buying, and {eta} is the Earth Mover's (Wasserstein-1) distance between p and p-hat. Note that when {eta} < o(sqrt(b)) this gives additive loss less than b (the trivial bound), and when {eta} is arbitrarily large (corresponding to an extremely inaccurate prediction) we still do not pay more than O(b log b) additive loss. An implication of these bounds is that our algorithm has consistency O(sqrt(b)) (additive loss when the prediction error is 0) and robustness O(b log b) (additive loss when the prediction error is arbitrarily large). Moreover, we do not need to assume that we know (or have any bound on) the prediction error {eta}, in contrast with previous work in robust optimization which assumes that we know this error. We complement this upper bound with a variety of lower bounds showing that it is essentially tight: not only can the consistency/robustness tradeoff not be improved, but our particular loss function cannot be meaningfully improved.

cs.LG↗

arXiv2025

Learning-Augmented Online Algorithms for Nonclairvoyant Joint Replenishment Problem with Deadlines

This paper considers using predictions in the context of the online Joint Replenishment Problem with Deadlines (JRP-D). Prior work includes asymptotically optimal competitive ratios of $O(1)$ for the clairvoyant setting and $O(\sqrt{n})$ of the nonclairvoyant setting, where $n$ is the number of items. The goal of this paper is to significantly reduce the competitive ratio for the nonclairvoyant case by leveraging predictions: when a request arrives, the true deadline of the request is not revealed, but the algorithm is given a predicted deadline. The main result is an algorithm whose competitive ratio is $O(\min(\eta^{1/3}\log^{2/3}(n), \sqrt{\eta}, \sqrt{n}))$, where $n$ is the number of item types and $\eta\leq n^2$ quantifies how flawed the predictions are in terms of the number of ``instantaneous item inversions.'' Thus, the algorithm is robust, i.e., it is never worse than the nonclairvoyant solution, and it is consistent, i.e., if the predictions exhibit no inversions, then the algorithm behaves similarly to the clairvoyant algorithm. Moreover, if the error is not too large, specifically $\eta < o(n^{3/2}/\log^2(n))$, then the algorithm obtains an asymptotically better competitive ratio than the nonclairvoyant algorithm. We also show that all deterministic algorithms falling in a certain reasonable class of algorithms have a competitive ratio of $\Omega(\eta^{1/3})$, so this algorithm is nearly the best possible with respect to this error metric.

cs.DS↗

arXiv2025

Controlling tail risk in two-slope ski rental

We study the optimal solution to a general two-slope ski rental problem with a tail risk, i.e., the chance of the competitive ratio exceeding a value $\gamma$ is bounded by $\delta$. This extends the recent study of tail bounds for ski rental by [Dinitz et al. SODA 2024] to the two-slope version defined by [Lotker et al. IPL 2008]. In this version, even after "buying" we must still pay a rental cost at each time step, though it is lower after buying. This models many real-world "rent-or-buy" scenarios where a one-time investment decreases (but does not eliminate) the per-time cost. Despite this being a simple extension of the classical problem, we find that adding tail risk bounds creates a fundamentally different solution structure. For example, in our setting there is a possibility that we never buy in an optimal solution (which can also occur without tail bounds), but more strangely (and unlike the case without tail bounds or the classical case with tail bounds) we also show that the optimal solution might need to have nontrivial probabilities of buying even at finite points beyond the time corresponding to the buying cost. Moreover, in many regimes there does not exist a unique optimal solution. As our first contribution, we develop a series of structure theorems to characterize some features of optimal solutions. The complex structure of optimal solutions makes it more difficult to develop an algorithm to compute such a solution. As our second contribution, we utilize our structure theorems to design two algorithms: one based on a greedy algorithm combined with binary search that is fast but yields arbitrarily close to optimal solutions, and a slower algorithm based on linear programming which computes exact optimal solutions.

cs.DS↗

arXiv2025

Minimum Congestion Routing of Unsplittable Flows in Data-Center Networks

Millions of flows are routed concurrently through a modern data-center. These networks are often built as Clos topologies, and flow demands are constrained only by the link capacities at the ingress and egress points. The minimum congestion routing problem seeks to route a set of flows through a data center while minimizing the maximum flow demand on any link. This is easily achieved by splitting flow demands along all available paths. However, arbitrary flow splitting is unrealistic. Instead, network operators rely on heuristics for routing unsplittable flows, the best of which results in a worst-case congestion of $2$ (twice the uniform link capacities). But is $2$ the lowest possible congestion? If not, can an efficient routing algorithm attain congestion below $2$? Guided by these questions, we investigate the minimum congestion routing problem in Clos networks with unsplittable flows. First, we show that for some sets of flows the minimum congestion is at least $\nicefrac{3}{2}$, and that it is $NP$-hard to approximate a minimum congestion routing by a factor less than $\nicefrac{3}{2}$. Second, addressing the motivating questions directly, we present a polynomial-time algorithm that guarantees a congestion of at most $\nicefrac{9}{5}$ for any set of flows, while also providing a $\nicefrac{9}{5}$ approximation of a minimum congestion routing. Last, shifting to the online setting, we demonstrate that no online algorithm (even randomized) can approximate a minimum congestion routing by a factor less than $2$, providing a strict separation between the online and the offline setting.

cs.NI↗

arXiv2025

A Generalized Binary Tree Mechanism for Differentially Private Approximation of All-Pair Distances

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 \textit{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 \textit{recursively separable}.

cs.DS↗

arXiv2025

Light Edge Fault Tolerant Graph Spanners

There has recently been significant interest in fault tolerant spanners, which are spanners that still maintain their stretch guarantees after some nodes or edges fail. This work has culminated in an almost complete understanding of the three-way tradeoff between stretch, sparsity, and number of faults tolerated. However, despite some progress in metric settings, there have been no results to date on the tradeoff in general graphs between stretch, lightness, and number of faults tolerated. We initiate the study of light edge fault tolerant (EFT) graph spanners, obtaining the first such results. First, we observe that lightness can be unbounded if we use the traditional definition (normalizing by the MST). We then argue that a natural definition of fault-tolerant lightness is to instead normalize by a min-weight fault tolerant connectivity preserver; essentially, a fault-tolerant version of the MST. However, even with this, we show that it is still not generally possible to construct $f$-EFT spanners whose weight compares reasonably to the weight of a min-weight $f$-EFT connectivity preserver. In light of this lower bound, it is natural to then consider bicriteria notions of lightness, where we compare the weight of an $f$-EFT spanner to a min-weight $(f' > f)$-EFT connectivity preserver. The most interesting question is to determine the minimum value of $f'$ that allows for reasonable lightness upper bounds. Our main result is a precise answer to this question: $f' = 2f$. In particular, we show that the lightness can be untenably large (roughly $n/k$ for a $k$-spanner) if one normalizes by the min-weight $(2f-1)$-EFT connectivity preserver. But if one normalizes by the min-weight $2f$-EFT connectivity preserver, then we show that the lightness is bounded by just $O(f^{1/2})$ times the non-fault tolerant lightness (roughly $n^{1/k}$, for a $(1+\epsilon)(2k-1)$-spanner).

cs.DS↗

arXiv2025

Approximation Algorithms for Optimal Hopsets

For a given graph $G$, a "hopset" $H$ with hopbound $\beta$ and stretch $\alpha$ is a set of edges such that between every pair of vertices $u$ and $v$, there is a path with at most $\beta$ hops in $G \cup H$ that approximates the distance between $u$ and $v$ up to a multiplicative stretch of $\alpha$. Hopsets have found a wide range of applications for distance-based problems in various computational models since the 90s. More recently, there has been significant interest in understanding these fundamental objects from an existential and structural perspective. But all of this work takes a worst-case (or existential) point of view: How many edges do we need to add to satisfy a given hopbound and stretch requirement for any input graph? We initiate the study of the natural optimization variant of this problem: given a specific graph instance, what is the minimum number of edges that satisfy the hopbound and stretch requirements? We give approximation algorithms for a generalized hopset problem which, when combined with known existential bounds, lead to different approximation guarantees for various regimes depending on hopbound, stretch, and directed vs. undirected inputs. We complement our upper bounds with a lower bound that implies Label Cover hardness for directed hopsets and shortcut sets with hopbound at least $3$.

cs.DS↗

arXiv2025

Differentially Private Matchings

Computing matchings in graphs is a foundational algorithmic task. Despite extensive interest in differentially private (DP) graph analysis, work on privately computing matching solutions, rather than just their size, has been sparse. The sole prior work in the standard model of pure $\varepsilon$-differential privacy, by Hsu, Huang, Roth, Roughgarden, and Wu [HHR+14, STOC'14], focused on allocations and was thus restricted to bipartite graphs. We present a comprehensive study of DP algorithms for maximum matching and b-matching in general graphs, which also yields techniques that improve upon the bipartite setting. En route to solving these matching problems, we develop a set of novel techniques with broad applicability, including a new symmetry argument for DP lower bounds, the first arboricity-based sparsifiers for node-DP, and the novel Public Vertex Subset Mechanism.

cs.DS↗

arXiv2024

Binary Search with Distributional Predictions

Algorithms with (machine-learned) predictions is a powerful framework for combining traditional worst-case algorithms with modern machine learning. However, the vast majority of work in this space assumes that the prediction itself is non-probabilistic, even if it is generated by some stochastic process (such as a machine learning system). This is a poor fit for modern ML, particularly modern neural networks, which naturally generate a distribution. We initiate the study of algorithms with distributional predictions, where the prediction itself is a distribution. We focus on one of the simplest yet fundamental settings: binary search (or searching a sorted array). This setting has one of the simplest algorithms with a point prediction, but what happens if the prediction is a distribution? We show that this is a richer setting: there are simple distributions where using the classical prediction-based algorithm with any single prediction does poorly. Motivated by this, as our main result, we give an algorithm with query complexity $O(H(p) + \log \eta)$, where $H(p)$ is the entropy of the true distribution $p$ and $\eta$ is the earth mover's distance between $p$ and the predicted distribution $\hat p$. This also yields the first distributionally-robust algorithm for the classical problem of computing an optimal binary search tree given a distribution over target keys. We complement this with a lower bound showing that this query complexity is essentially optimal (up to constants), and experiments validating the practical usefulness of our algorithm.

cs.LG↗

arXiv2024

Noise is All You Need: Private Second-Order Convergence of Noisy SGD

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.

cs.LG↗

arXiv2024

Parallel Set Cover and Hypergraph Matching via Uniform Random Sampling

The SetCover problem has been extensively studied in many different models of computation, including parallel and distributed settings. From an approximation point of view, there are two standard guarantees: an $O(\log \Delta)$-approximation (where $\Delta$ is the maximum set size) and an $O(f)$-approximation (where $f$ is the maximum number of sets containing any given element). In this paper, we introduce a new, surprisingly simple, model-independent approach to solving SetCover in unweighted graphs. We obtain multiple improved algorithms in the MPC and CRCW PRAM models. First, in the MPC model with sublinear space per machine, our algorithms can compute an $O(f)$ approximation to SetCover in $\hat{O}(\sqrt{\log \Delta} + \log f)$ rounds, where we use the $\hat{O}(x)$ notation to suppress $\mathrm{poly} \log x$ and $\mathrm{poly} \log \log n$ terms, and a $O(\log \Delta)$ approximation in $O(\log^{3/2} n)$ rounds. Moreover, in the PRAM model, we give a $O(f)$ approximate algorithm using linear work and $O(\log n)$ depth. All these bounds improve the existing round complexity/depth bounds by a $\log^{\Omega(1)} n$ factor. Moreover, our approach leads to many other new algorithms, including improved algorithms for the HypergraphMatching problem in the MPC model, as well as simpler SetCover algorithms that match the existing bounds.

cs.DS↗

arXiv2024

Differentially Private Algorithms for Graph Cuts: A Shifting Mechanism Approach and More

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 $O(k\log n)$ (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 $\tilde{O}(k^{1.5})$.

cs.CR↗

arXiv2024

Approximation Algorithms for Minimizing Congestion in Demand-Aware Networks

Emerging reconfigurable optical communication technologies allow to enhance datacenter topologies with demand-aware links optimized towards traffic patterns. This paper studies the algorithmic problem of jointly optimizing topology and routing in such demand-aware networks to minimize congestion, along two dimensions: (1) splittable or unsplittable flows, and (2) whether routing is segregated, i.e., whether routes can or cannot combine both demand-aware and demand-oblivious (static) links. For splittable and segregated routing, we show that the problem is generally $2$-approximable, but APX-hard even for uniform demands induced by a bipartite demand graph. For unsplittable and segregated routing, we establish upper and lower bounds of $O\left(\log m/ \log\log m \right)$ and $\Omega\left(\log m/ \log\log m \right)$, respectively, for polynomial-time approximation algorithms, where $m$ is the number of static links. We further reveal that under un-/splittable and non-segregated routing, even for demands of a single source (resp., destination), the problem cannot be approximated better than $\Omega\left(\frac{c_{\max}}{c_{\min}} \right)$ unless P=NP, where $c_{\max}$ (resp., $c_{\min}$) denotes the maximum (resp., minimum) capacity. It remains NP-hard for uniform capacities, but is tractable for a single commodity and uniform capacities. Our trace-driven simulations show a significant reduction in network congestion compared to existing solutions.

cs.PF↗

arXiv2023

Almost Tight Bounds for Differentially Private Densest Subgraph

We study the Densest Subgraph (DSG) problem under the additional constraint of differential privacy. DSG is a fundamental theoretical question which plays a central role in graph analytics, and so privacy is a natural requirement. All known private algorithms for Densest Subgraph lose constant multiplicative factors, despite the existence of non-private exact algorithms. We show that, perhaps surprisingly, this loss is not necessary: in both the classic differential privacy model and the LEDP model (local edge differential privacy, introduced recently by Dhulipala et al. [FOCS 2022]), we give $(\epsilon, \delta)$-differentially private algorithms with no multiplicative loss whatsoever. In other words, the loss is \emph{purely additive}. Moreover, our additive losses match or improve the best-known previous additive loss (in any version of differential privacy) when $1/\delta$ is polynomial in $n$, and are almost tight: in the centralized setting, our additive loss is $O(\log n /\epsilon)$ while there is a known lower bound of $\Omega(\sqrt{\log n / \epsilon})$. We also give a number of extensions. First, we show how to extend our techniques to both the node-weighted and the directed versions of the problem. Second, we give a separate algorithm with pure differential privacy (as opposed to approximate DP) but with worse approximation bounds. And third, we give a new algorithm for privately computing the optimal density which implies a separation between the structural problem of privately computing the densest subgraph and the numeric problem of privately computing the density of the densest subgraph.

cs.DS↗

arXiv2023

Controlling Tail Risk in Online Ski-Rental

The classical ski-rental problem admits a textbook 2-competitive deterministic algorithm, and a simple randomized algorithm that is $\frac{e}{e-1}$-competitive in expectation. The randomized algorithm, while optimal in expectation, has a large variance in its performance: it has more than a 37% chance of competitive ratio exceeding 2, and a $\Theta(1/n)$ chance of the competitive ratio exceeding $n$! We ask what happens to the optimal solution if we insist that the tail risk, i.e., the chance of the competitive ratio exceeding a specific value, is bounded by some constant $\delta$. We find that this additional modification significantly changes the structure of the optimal solution. The probability of purchasing skis on a given day becomes non-monotone, discontinuous, and arbitrarily large (for sufficiently small tail risk $\delta$ and large purchase cost $n$).

cs.DS↗

arXiv2023

Improved Approximations for Relative Survivable Network Design

One of the most important and well-studied settings for network design is edge-connectivity requirements. This encompasses uniform demands such as the Minimum $k$-Edge-Connected Spanning Subgraph problem as well as nonuniform demands such as the Survivable Network Design problem (SND). In a recent paper by [Dinitz, Koranteng, Kortsarz APPROX '22] , the authors observed that a weakness of these formulations is that it does not enable one to consider fault-tolerance in graphs that have just a few small cuts. To remedy this, they introduced new variants of these problems under the notion "relative" fault-tolerance. Informally, this requires not that two nodes are connected if there are a bounded number of faults (as in the classical setting), but that two nodes are connected if there are a bounded number of faults and the two nodes are connected in the underlying graph post-faults. The problem is already highly non-trivial even for the case of a single demand. Due to difficulties introduced by this new notion of fault-tolerance, the results in [Dinitz, Koranteng, Kortsarz APPROX '22] are quite limited. For the Relative Survivable Network Design problem (RSND), when the demands are not uniform they give a nontrivial result only when there is a single demand with a connectivity requirement of $3$: a non-optimal $27/4$-approximation. We strengthen this result in two significant ways: We give a $2$-approximation for RSND where all requirements are at most $3$, and a $2^{O(k^2)}$-approximation for RSND with a single demand of arbitrary value $k$. To achieve these results, we first use the "cactus representation'' of minimum cuts to give a lossless reduction to normal SND. Second, we extend the techniques of [Dinitz, Koranteng, Kortsarz APPROX '22] to prove a generalized and more complex version of their structure theorem, which we then use to design a recursive approximation algorithm.

cs.DS↗

arXiv2023

Degrees and Network Design: New Problems and Approximations

While much of network design focuses mostly on cost (number or weight of edges), node degrees have also played an important role. They have traditionally either appeared as an objective, to minimize the maximum degree (e.g., the Minimum Degree Spanning Tree problem), or as constraints which might be violated to give bicriteria approximations (e.g., the Minimum Cost Degree Bounded Spanning Tree problem). We extend the study of degrees in network design in two ways. First, we introduce and study a new variant of the Survivable Network Design Problem where in addition to the traditional objective of minimizing the cost of the chosen edges, we add a constraint that the $\ell_p$-norm of the node degree vector is bounded by an input parameter. This interpolates between the classical settings of maximum degree (the $\ell_{\infty}$-norm) and the number of edges (the $\ell_1$-degree), and has natural applications in distributed systems and VLSI design. We give a constant bicriteria approximation in both measures using convex programming. Second, we provide a polylogrithmic bicriteria approximation for the Degree Bounded Group Steiner problem on bounded treewidth graphs, solving an open problem from [Kortsarz and Nutov, Discret. Appl. Math. 2022] and [Guo et al., Algorithmica 2022].

cs.DS↗