arXiv subjects

Quanquan C. Liu

Publications and source records attributed to Quanquan C. Liu.

At least 19 recordsLinked to original sources

arXiv2026

QEDBENCH: Quantifying the Alignment Gap in Automated Evaluation of University-Level Mathematical Proofs

As Large Language Models (LLMs) saturate elementary benchmarks, the research frontier has shifted from generation to the reliability of automated evaluation. We demonstrate that standard "LLM-as-a-Judge" protocols suffer from a systematic Alignment Gap when applied to upper-undergraduate to early graduate level mathematics. To quantify this, we introduce QEDBench, the first large-scale dual-rubric alignment benchmark to systematically measure alignment with human experts on university-level math proofs by contrasting course-specific rubrics against expert common knowledge criteria. By deploying a dual-evaluation matrix (7 judges x 5 solvers) against 1,000+ hours of human evaluation, we reveal that certain frontier evaluators like Claude Opus 4.5, DeepSeek-V3, Qwen 2.5 Max, and Llama 4 Maverick exhibit significant positive bias (up to +0.18, +0.20, +0.30, +0.36 mean score inflation, respectively). Furthermore, we uncover a critical reasoning gap in the discrete domain: while Gemini 3.0 Pro achieves state-of-the-art performance (0.91 average human evaluation score), other reasoning models like GPT-5 Pro and Claude Sonnet 4.5 see their performance significantly degrade in discrete domains. Specifically, their average human evaluation scores drop to 0.72 and 0.63 in Discrete Math, and to 0.74 and 0.50 in Graph Theory. In addition to these research results, we also release QEDBench as a public benchmark for evaluating and improving AI judges. Our benchmark is publicly published at https://github.com/qqliu/Yale-QEDBench.

cs.LG↗

arXiv2026

ParEVO: Synthesizing Code for Irregular Data: High-Performance Parallelism through Agentic Evolution

The transition from sequential to parallel computing is essential for modern high-performance applications but is hindered by the steep learning curve of concurrent programming. This challenge is magnified for irregular data structures (such as sparse graphs, unbalanced trees, and non-uniform meshes) where static scheduling fails and data dependencies are unpredictable. Current Large Language Models (LLMs) often fail catastrophically on these tasks, generating code plagued by subtle race conditions, deadlocks, and sub-optimal scaling. We bridge this gap with ParEVO, a framework designed to synthesize high-performance parallel algorithms for irregular data. Our contributions include: (1) The Parlay-Instruct Corpus, a curated dataset of 13,820 tasks synthesized via a "Critic-Refine" pipeline that explicitly filters for empirically performant algorithms that effectively utilize Work-Span parallel primitives; (2) specialized DeepSeek, Qwen, and Gemini models fine-tuned to align probabilistic generation with the rigorous semantics of the ParlayLib library; and (3) an Evolutionary Coding Agent (ECA) that improves the "last mile" of correctness by iteratively repairing code using feedback from compilers, dynamic race detectors, and performance profilers. On the ParEval benchmark, ParEVO achieves an average 106x speedup (with a maximum of 1103x) across the suite, and a robust 13.6x speedup specifically on complex irregular graph problems, outperforming state-of-the-art commercial models. Furthermore, our evolutionary approach matches state-of-the-art expert human baselines, achieving up to a 4.1x speedup on specific highly-irregular kernels. Source code and datasets are available at https://github.com/WildAlg/ParEVO.

cs.LG↗

arXiv2026

Reasoning with Sampling: Cutting at Decision Points

Frontier reasoning models are produced by posttraining base language models with reinforcement learning. Recent work has challenged this by showing that sampling from a sharpened version of the base model's distribution, a so-called power distribution, elicits comparable reasoning without additional training, curated datasets, or verifiers. However, making this method practical requires efficiently sampling from the power distribution. A sampler needs to "mix" to the power distribution, which necessitates moving between modes of the target distribution; intuitively, e.g., trying different reasoning strategies. The samplers proposed in prior works repeatedly select a "cut" position in the current reasoning trace uniformly at random and resample the suffix from that position onward. However, reasoning traces typically contain a few consequential decisions (e.g., the choice of proof strategy or algorithm), and we observe that a uniformly chosen cut tends to rewrite local details rather than revisit decision points. We introduce an algorithm (Entropy-Cut Metropolis-Hastings) that uses the base model's next-token entropy as a proxy to identify key decision points and resample from those positions. We empirically verify that entropy jumps are a useful proxy for decision points and, in a stylized model of reasoning, prove that our method's mixing time scales with the number of decisions in a trace rather than with the number of tokens, which can be much larger. Across MATH500, HumanEval, GPQA Diamond, and AIME26, our method consistently improves over baselines and RL-trained models.

cs.LG↗

arXiv2026

LAPRAS : Learning-Augmented PRivate Answering for linear query Streams

Modern database workloads are highly predictable: query streams are dominated by recurring jobs and templates, even when their arrival order is not known in advance. This motivates a learning-augmented view of online differentially private (DP) analytics: can algorithms utilize predictions about which queries will occur to improve utility under a single global privacy budget, while remaining robust when predictions are wrong? We study online DP query answering, where a curator must answer a stream $Q$ of $S$ linear queries arriving in uniformly random order under privacy budget $(ε,δ)$. We present LAPRAS, which assumes access to an oracle that outputs a prediction set of queries likely to appear in the stream and uses it to guide privacy spending. LAPRAS answers predicted queries using the offline-optimal Matrix Mechanism and answers the remaining queries online from a residual budget. To pace spending across an unknown number of unpredicted queries, we introduce Smooth Allocation, which forms an unbiased stopping-time estimate $\widehat{B}$ from the first $T=Θ(\log^2 S)$ unpredicted queries and continuously recalibrates per-query expenditure. Empirically, over two real datasets, we validate the intended consistency--robustness trade-off: LAPRAS achieves near-offline utility under high overlap and degrades gracefully to baseline-level performance when overlap is low.

cs.CR↗

arXiv2026

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↗

arXiv2025

Pointwise Lipschitz Continuous Graph Algorithms

In many real-world applications, it is undesirable to drastically change the problem solution after a small perturbation in the input, as unstable outputs can lead to costly transaction fees, privacy and security concerns, reduced user trust, and lack of replicability. Despite the widespread application of graph algorithms, many classical algorithms are not robust to small input disturbances. Towards addressing this issue, we study the pointwise Lipschitz continuity of graph algorithms, a notion of stability introduced by Kumabe and Yoshida [KY23, FOCS'23] and further studied in related settings [KY24, ICALP'24], [KY25, SODA'25], [GKY25, ESA'25]. Our main result is a linear programming (LP) based minimum $S$-$T$ cut algorithm with a provably optimal Lipschitz constant, as witnessed by an accompanying lower bound. As a direct corollary, we give the first dynamic minimum $S$-$T$ cut algorithm with non-trivial recourse bound. At the core of our techniques is a novel framework for analyzing the Lipschitz constant of regularized LP relaxations. Our framework crucially unlocks the use of weighted regularizers, which could not be analyzed through previous methods, and leads to polynomial improvements in the Lipschitz constant compared to what is achievable through previous techniques. To demonstrate the flexibility of our methods, we also design an LP-based $b$-matching algorithm that improves on the state-of-the-art [KY23] Lipschitz constant in certain input regimes when $b\equiv 1$. Moreover, our algorithm cleanly extends to the general case when $b\geq 1$, whereas [KY23] is specialized to the case of $b\equiv 1$.

cs.DS↗

arXiv2025

Practical and Accurate Local Edge Differentially Private Graph Algorithms

The rise of massive networks across diverse domains necessitates sophisticated graph analytics, often involving sensitive data and raising privacy concerns. This paper addresses these challenges using local differential privacy (LDP), which enforces privacy at the individual level, where no third-party entity is trusted, unlike centralized models that assume a trusted curator. We introduce novel LDP algorithms for two fundamental graph statistics: k-core decomposition and triangle counting. Our approach leverages input-dependent private graph properties, specifically the degeneracy and maximum degree of the graph, to improve theoretical utility. Unlike prior methods, our error bounds are determined by the maximum degree rather than the total number of edges, resulting in significantly tighter guarantees. For triangle counting, we improve upon the work of Imola, Murakami, and Chaudhury [USENIX Security `21, `22], which bounds error in terms of edge count. Instead, our algorithm achieves bounds based on graph degeneracy by leveraging a private out-degree orientation, a refined variant of Eden et al.'s randomized response technique [ICALP `23], and a novel analysis, yielding stronger guarantees than prior work. Beyond theoretical gains, we are the first to evaluate local DP algorithms in a distributed simulation, unlike prior work tested on a single processor. Experiments on real-world graphs show substantial accuracy gains: our k-core decomposition achieves errors within 3x of exact values, far outperforming the 131x error in the baseline of Dhulipala et al. [FOCS `22]. Our triangle counting algorithm reduces multiplicative approximation errors by up to six orders of magnitude, while maintaining competitive runtime.

cs.DS↗

arXiv2025

Triangle Counting with Local Edge Differential Privacy

Many deployments of differential privacy in industry are in the local model, where each party releases its private information via a differentially private randomizer. We study triangle counting in the local model with edge differential privacy (that, intuitively, requires that the outputs of the algorithm on graphs that differ in one edge be indistinguishable). In this model, each party's local view consists of the adjacency list of one vertex. We investigate both noninteractive and interactive variants of the model. In the noninteractive model, we prove that additive $Ω(n^2)$ error is necessary for sufficiently small constant $\varepsilon$, where $n$ is the number of nodes and $\varepsilon$ is the privacy parameter. This lower bound is our main technical contribution. It uses a reconstruction attack with a new class of linear queries and a novel mix-and-match strategy of running the local randomizers with different completions of their adjacency lists. It matches the additive error of the algorithm based on Randomized Response, proposed by Imola, Murakami and Chaudhuri (USENIX2021) and analyzed by Imola, Murakami and Chaudhuri (CCS2022) for constant $\varepsilon$. We use a different postprocessing of Randomized Response and provide tight bounds on the variance of the resulting algorithm. In the interactive setting, we prove a lower bound of $Ω(n^{3/2}/\varepsilon)$ on the additive error for $\varepsilon\leq 1$. Previously, no hardness results were known for interactive, edge-private algorithms in the local model, except for those that follow trivially from the results for the central model. Our work significantly improves on the state of the art in differentially private graph analysis in the local model.

cs.DS↗

arXiv2025

Near-Optimal Differentially Private Graph Algorithms via the Multidimensional AboveThreshold Mechanism

Many differentially private and classical non-private graph algorithms rely crucially on determining whether some property of each vertex meets a threshold. For example, for the $k$-core decomposition problem, the classic peeling algorithm iteratively removes a vertex if its induced degree falls below a threshold. The sparse vector technique (SVT) is generally used to transform non-private threshold queries into private ones with only a small additive loss in accuracy. However, a naive application of SVT in the graph setting leads to an amplification of the error by a factor of $n$ due to composition, as SVT is applied to every vertex. In this paper, we resolve this problem by formulating a novel generalized sparse vector technique which we call the Multidimensional AboveThreshold (MAT) Mechanism which generalizes SVT (applied to vectors with one dimension) to vectors with multiple dimensions. As an application, we solve a number of important graph problems with better bounds than previous work. We apply our MAT mechanism to obtain a set of improved bounds for a variety of problems including $k$-core decomposition, densest subgraph, low out-degree ordering, and vertex coloring. We give a tight local edge DP algorithm for $k$-core decomposition with $O(ε^{-1}\log n)$ additive error and no multiplicative error in $O(n)$ rounds. We also give a new $(2+η)$-factor multiplicative, $O(ε^{-1}\log n)$ additive error algorithm in $O(\log^2 n)$ rounds for any constant $η> 0$. Both of these results are asymptotically tight against our new lower bound of $Ω(\log n)$ for any constant-factor approximation algorithm for $k$-core decomposition. Our new algorithms for $k$-core also directly lead to new algorithms for densest subgraph and low out-degree ordering. Our novel private defective coloring algorithms uses number of colors proportional to the arboricity of the graph.

cs.DS↗

arXiv2025

Sublinear Space Graph Algorithms in the Continual Release Model

The graph continual release model of differential privacy seeks to produce differentially private solutions to graph problems under a stream of edge updates where new private solutions are released after each update. Thus far, previously known edge-differentially private algorithms for most graph problems including densest subgraph and matchings in the continual release setting only output real-value estimates (not vertex subset solutions) and do not use sublinear space. Instead, they rely on computing exact graph statistics on the input [FHO21,SLMVC18]. In this paper, we leverage sparsification to address the above shortcomings for edge-insertion streams. Our edge-differentially private algorithms use sublinear space with respect to the number of edges in the graph while some also achieve sublinear space in the number of vertices in the graph. In addition, for the densest subgraph problem, we also output edge-differentially private vertex subset solutions; no previous graph algorithms in the continual release model output such subsets. We make novel use of assorted sparsification techniques from the non-private streaming and static graph algorithms literature to achieve new results in the sublinear space, continual release setting. This includes algorithms for densest subgraph, maximum matching, as well as the first continual release $k$-core decomposition algorithm. We also develop a novel sparse level data structure for $k$-core decomposition that may be of independent interest. To complement our insertion-only algorithms, we conclude with polynomial additive error lower bounds for edge-privacy in the fully dynamic setting, where only logarithmic lower bounds were previously known.

cs.DS↗

arXiv2024

Improved Massively Parallel Triangle Counting in $O(1)$ Rounds

In this short note, we give a novel algorithm for $O(1)$ round triangle counting in bounded arboricity graphs. Counting triangles in $O(1)$ rounds (exactly) is listed as one of the interesting remaining open problems in the recent survey of Im et al. [IKLMV23]. The previous paper of Biswas et al. [BELMR20], which achieved the best bounds under this setting, used $O(\log \log n)$ rounds in sublinear space per machine and $O(mα)$ total space where $α$ is the arboricity of the graph and $n$ and $m$ are the number of vertices and edges in the graph, respectively. Our new algorithm is very simple, achieves the optimal $O(1)$ rounds without increasing the space per machine and the total space, and has the potential of being easily implementable in practice.

cs.DS↗

arXiv2024

Near-Optimal Differentially Private k-Core Decomposition

Recent work by Dhulipala et al. \cite{DLRSSY22} initiated the study of the $k$-core decomposition problem under differential privacy via a connection between low round/depth distributed/parallel graph algorithms and private algorithms with small error bounds. They showed that one can output differentially private approximate $k$-core numbers, while only incurring a multiplicative error of $(2 +η)$ (for any constant $η>0$) and additive error of $\poly(\log(n))/\eps$. In this paper, we revisit this problem. Our main result is an $\eps$-edge differentially private algorithm for $k$-core decomposition which outputs the core numbers with no multiplicative error and $O(\text{log}(n)/\eps)$ additive error. This improves upon previous work by a factor of 2 in the multiplicative error, while giving near-optimal additive error. Our result relies on a novel generalized form of the sparse vector technique, which is especially well-suited for threshold-based graph algorithms; thus, we further strengthen the connection between distributed/parallel graph algorithms and differentially private algorithms.

cs.DS↗

arXiv2024

Parallel $k$-Core Decomposition with Batched Updates and Asynchronous Reads

Maintaining a dynamic $k$-core decomposition is an important problem that identifies dense subgraphs in dynamically changing graphs. Recent work by Liu et al. [SPAA 2022] presents a parallel batch-dynamic algorithm for maintaining an approximate $k$-core decomposition. In their solution, both reads and updates need to be batched, and therefore each type of operation can incur high latency waiting for the other type to finish. To tackle most real-world workloads, which are dominated by reads, this paper presents a novel hybrid concurrent-parallel dynamic $k$-core data structure where asynchronous reads can proceed concurrently with batches of updates, leading to significantly lower read latencies. Our approach is based on tracking causal dependencies between updates, so that causally related groups of updates appear atomic to concurrent readers. Our data structure guarantees linearizability and liveness for both reads and updates, and maintains the same approximation guarantees as prior work. Our experimental evaluation on a 30-core machine shows that our approach reduces read latency by orders of magnitude compared to the batch-dynamic algorithm, up to a $\left(4.05 \cdot 10^{5}\right)$-factor. Compared to an unsynchronized (non-linearizable) baseline, our read latency overhead is only up to a $3.21$-factor greater, while improving accuracy of coreness estimates by up to a factor of $52.7$.

cs.DC↗

arXiv2023

The Predicted-Updates Dynamic Model: Offline, Incremental, and Decremental to Fully Dynamic Transformations

We formulate the predicted-updates dynamic model, one of the first beyond-worst-case models for dynamic algorithms, which generalizes a large set of well-studied dynamic models including the offline dynamic, incremental, and decremental models to the fully dynamic setting when given predictions about the update times of the elements. In the most basic form of our model, we receive a set of predicted update times for all of the updates that occur over the event horizon. We give a novel framework that "lifts" offline divide-and-conquer algorithms into the fully dynamic setting with little overhead. Using this, we are able to interpolate between the offline and fully dynamic settings; when the $\ell_1$ error of the prediction is linear in the number of updates, we achieve the offline runtime of the algorithm (up to $\mathrm{poly} \log n$ factors). Provided a fully dynamic backstop algorithm, our algorithm will never do worse than the backstop algorithm regardless of the prediction error. Furthermore, our framework achieves a smooth linear trade-off between $\ell_1$ error in the predictions and runtime. These correspond to the desiderata of consistency, robustness, and graceful degradation of the algorithms-with-predictions literature. We further extend our techniques to incremental and decremental settings, transforming algorithms in these settings when given predictions of only the deletion and insertion times, respectively. Our framework is general, and we apply it to obtain improved efficiency bounds over the state-of-the-art dynamic algorithms for a variety of problems including triconnectivity, planar digraph all pairs shortest paths, $k$-edge connectivity, and others, for prediction error of reasonable magnitude.

cs.DS↗

arXiv2023

Practical Parallel Algorithms for Near-Optimal Densest Subgraphs on Massive Graphs

The densest subgraph problem has received significant attention, both in theory and in practice, due to its applications in problems such as community detection, social network analysis, and spam detection. Due to the high cost of obtaining exact solutions, much attention has focused on designing approximate densest subgraph algorithms. However, existing approaches are not able to scale to massive graphs with billions of edges. In this paper, we introduce a new framework that combines approximate densest subgraph algorithms with a pruning optimization. We design new parallel variants of the state-of-the-art sequential Greedy++ algorithm, and plug it into our framework in conjunction with a parallel pruning technique based on $k$-core decomposition to obtain parallel $(1+\varepsilon)$-approximate densest subgraph algorithms. On a single thread, our algorithms achieve $2.6$--$34\times$ speedup over Greedy++, and obtain up to $22.37\times$ self relative parallel speedup on a 30-core machine with two-way hyper-threading. Compared with the state-of-the-art parallel algorithm by Harb et al. [NeurIPS'22], we achieve up to a $114\times$ speedup on the same machine. Finally, against the recent sequential algorithm of Xu et al. [PACMMOD'23], we achieve up to a $25.9\times$ speedup. The scalability of our algorithms enables us to obtain near-optimal density statistics on the hyperlink2012 (with roughly 113 billion edges) and clueweb (with roughly 37 billion edges) graphs for the first time in the literature.

cs.DS↗

arXiv2023

Parallel Batch-Dynamic Algorithms for $k$-Core Decomposition and Related Graph Problems

Maintaining a $k$-core decomposition quickly in a dynamic graph has important applications in network analysis. The main challenge for designing efficient exact algorithms is that a single update to the graph can cause significant global changes. Our paper focuses on \emph{approximation} algorithms with small approximation factors that are much more efficient than what exact algorithms can obtain. We present the first parallel, batch-dynamic algorithm for approximate $k$-core decomposition that is efficient in both theory and practice. Our algorithm is based on our novel parallel level data structure, inspired by the sequential level data structures of Bhattacharya et al [STOC '15] and Henzinger et al [2020]. Given a graph with $n$ vertices and a batch of updates $\mathcal{B}$, our algorithm provably maintains a $(2 + \varepsilon)$-approximation of the coreness values of all vertices (for any constant $\varepsilon > 0$) in $O(|\mathcal{B}|\log^2 n)$ amortized work and $O(\log^2 n \log\log n)$ depth (parallel time) with high probability. As a by-product, our $k$-core decomposition algorithm also gives a batch-dynamic algorithm for maintaining an $O(α)$ out-degree orientation, where $α$ is the current arboricity of the graph. We demonstrate the usefulness of our low out-degree orientation algorithm by presenting a new framework to formally study batch-dynamic algorithms in bounded-arboricity graphs. Our framework obtains new provably-efficient parallel batch-dynamic algorithms for maximal matching, clique counting, and vertex coloring. We implemented and experimentally evaluated our $k$-core decomposition algorithm on a 30-core machine with two-way hyper-threading on $11$ graphs of varying densities and sizes. [...]

cs.DS↗

arXiv2023

Scalable Auction Algorithms for Bipartite Maximum Matching Problems

In this paper, we give new auction algorithms for maximum weighted bipartite matching (MWM) and maximum cardinality bipartite $b$-matching (MCbM). Our algorithms run in $O\left(\log n/\varepsilon^8\right)$ and $O\left(\log n/\varepsilon^2\right)$ rounds, respectively, in the blackboard distributed setting. We show that our MWM algorithm can be implemented in the distributed, interactive setting using $O(\log^2 n)$ and $O(\log n)$ bit messages, respectively, directly answering the open question posed by Demange, Gale and Sotomayor [DNO14]. Furthermore, we implement our algorithms in a variety of other models including the the semi-streaming model, the shared-memory work-depth model, and the massively parallel computation model. Our semi-streaming MWM algorithm uses $O(1/\varepsilon^8)$ passes in $O(n \log n \cdot \log(1/\varepsilon))$ space and our MCbM algorithm runs in $O(1/\varepsilon^2)$ passes using $O\left(\left(\sum_{i \in L} b_i + |R|\right)\log(1/\varepsilon)\right)$ space (where parameters $b_i$ represent the degree constraints on the $b$-matching and $L$ and $R$ represent the left and right side of the bipartite graph, respectively). Both of these algorithms improves \emph{exponentially} the dependence on $\varepsilon$ in the space complexity in the semi-streaming model against the best-known algorithms for these problems, in addition to improvements in round complexity for MCbM. Finally, our algorithms eliminate the large polylogarithmic dependence on $n$ in depth and number of rounds in the work-depth and massively parallel computation models, respectively, improving on previous results which have large polylogarithmic dependence on $n$ (and exponential dependence on $\varepsilon$ in the MPC model).

cs.DS↗

arXiv2023

An Algorithmic Approach to Address Course Enrollment Challenges

Massive surges of enrollments in courses have led to a crisis in several computer science departments - not only is the demand for certain courses extremely high from majors, but the demand from non-majors is also very high. Much of the time, this leads to significant frustration on the part of the students, and getting seats in desired courses is a rather ad-hoc process. One approach is to first collect information from students about which courses they want to take and to develop optimization models for assigning students to available seats in a fair manner. What makes this problem complex is that the courses themselves have time conflicts, and the students have credit caps (an upper bound on the number of courses they would like to enroll in). We model this problem as follows. We have $n$ agents (students), and there are "resources" (these correspond to courses). Each agent is only interested in a subset of the resources (courses of interest), and each resource can only be assigned to a bounded number of agents (available seats). In addition, each resource corresponds to an interval of time, and the objective is to assign non-overlapping resources to agents so as to produce "fair and high utility" schedules. In this model, we provide a number of results under various settings and objective functions. Specifically, in this paper, we consider the following objective functions: total utility, max-min (Santa Claus objective), and envy-freeness. The total utility objective function maximizes the sum of the utilities of all courses assigned to students. The max-min objective maximizes the minimum utility obtained by any student. Finally, envy-freeness ensures that no student envies another student's allocation. Under these settings and objective functions, we show a number of theoretical results. Specifically, we show that the course allocation under [...]

cs.DS↗