arXiv (Cornell University)最新文献

MEMONDEMAND: A Memory Management System for Large-Scale Enterprise Data MEMONDEMAND：大型企业数据的内存管理系统

arXiv (Cornell University) Pub Date : 2026-08-22 DOI: 10.48550/arxiv.2608.22141

Xinyuan Song, Bowen Zhu, Hasibul Haque, Liang Zhao

{"title":"MEMONDEMAND: A Memory Management System for Large-Scale Enterprise Data","authors":"Xinyuan Song, Bowen Zhu, Hasibul Haque, Liang Zhao","doi":"10.48550/arxiv.2608.22141","DOIUrl":"https://doi.org/10.48550/arxiv.2608.22141","url":null,"abstract":"Enterprise repositories are large, heteroge- neous, and continuously updated, making re- trieval difficult when efficient access, source- faithful evidence, and cross-query adaptation must be supported together. Enterprise mem- ory extends retrieval beyond the model con- text, but existing systems do not jointly address collection-specific hierarchy construction, low- cost routing, detailed evidence loading, and workload-aware memory updates at this scale. We introduce MEMONDEMAND, short for On- Demand Memory, a memory management sys- tem with three coordinated mechanisms: a dy- namic multi-level hierarchy that determines the abstraction structure and depth for each col- lection, dual memory at every hierarchy level that separates distilled routing from detailed evidence, and on-demand memory promotion that updates node priority under a bounded active-state budget. On EnterpriseRAG-Bench, MEMONDEMAND outperforms the strongest published LB#1 result at every evaluated scale from 10M tokens through the complete 618M- token collection, with gains of 12.23% at 10M and 4.66% at 618M. Results on FinanceBench, HotpotQA, and FRAMES further show strong performance across financial, multi-hop, and fact-retrieval settings. Together, these results establish MEMONDEMAND as an accurate, ef- ficient, and scalable memory solution for very large enterprise repositories across data scales, domains, and evidence requirements. Our code is available at https://github.com/ xfab-xinyuansong/MemOnDemand.git.","PeriodicalId":496270,"journal":{"name":"arXiv (Cornell University)","volume":"1 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-08-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://arxiv.org/pdf/2608.22141","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148861351","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Proof-Valid Caching under Premise Erasures: Local Structural Limits and Shared-Workload Gains 前提擦除下的证明有效缓存：本地结构限制和共享工作负载增益

arXiv (Cornell University) Pub Date : 2026-08-12 DOI: 10.48550/arxiv.2608.11782

Jianfeng Xu

{"title":"Proof-Valid Caching under Premise Erasures: Local Structural Limits and Shared-Workload Gains","authors":"Jianfeng Xu","doi":"10.48550/arxiv.2608.11782","DOIUrl":"https://doi.org/10.48550/arxiv.2608.11782","url":null,"abstract":"We study reliable query recovery under independent premise erasures in semantically transparent caching systems, where every cached object must be a logical consequence of the premise base. Recovery succeeds only when the query remains derivable from surviving premises and the cache. Under a deterministic canonical-witness regime, we prove a query-local projection theorem and an exact residual-leaf law: recovery fails exactly when an erased base leaf retains a cache-free path to the query. Single-query design becomes weighted partial path interception. For shared workloads, we introduce semantic modules and derive exact reliability laws under joint and maximal-error criteria. The shared-module cache is exactly optimal under exact module routing and homogeneous costs, whereas optimal selection in general derivation DAGs is NP-complete at depth two. Against a coded benchmark recovering workload-relevant leaf payloads, MDS parity caching is optimal up to one packet. Leaf-only transparency incurs a first-order overhead inversely proportional to the erasure rate; shared modules multiply that inverse-erasure-rate scaling by the module-to-leaf cost ratio divided by the number of protected leaves. A Datalog instance and Monte Carlo checks illustrate the theory. For derivation-structured content, the results provide exact stochastic-erasure counterparts of function-correcting storage and an exact distributional quantification of maximal recoverability.","PeriodicalId":496270,"journal":{"name":"arXiv (Cornell University)","volume":"1 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-08-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://arxiv.org/pdf/2608.11782","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148861348","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

PSP: Low-Overhead Packet-Level Load Balancing for Stale-State and Bandwidth-Asymmetric Networks 状态和带宽不对称网络的低开销包级负载均衡

arXiv (Cornell University) Pub Date : 2026-08-09 DOI: 10.48550/arxiv.2608.08425

Jiaqi Liu, Chunyang Zhang, Heng Pan, Yanbiao Li

{"title":"PSP: Low-Overhead Packet-Level Load Balancing for Stale-State and Bandwidth-Asymmetric Networks","authors":"Jiaqi Liu, Chunyang Zhang, Heng Pan, Yanbiao Li","doi":"10.48550/arxiv.2608.08425","DOIUrl":"https://doi.org/10.48550/arxiv.2608.08425","url":null,"abstract":"With the rapid growth of large language model training and generative artificial intelligence services, data center networks face severe micro-burst traffic and high concurrency. Traditional hash-based flow-level load balancing cannot sense link states, leading to hash collisions, hotspot congestion, and tail latency in multipath Clos networks. Existing packet-level schemes are constrained by stale state information, high hardware complexity, and poor adaptation to heterogeneous links. To address these issues, this paper proposes probabilistic state-proportional (PSP) dispatching, a packet-level load balancing algorithm. Using a Band-based discrete state representation, PSP replaces global sorting with local probability mapping, reducing hardware complexity while suppressing herding and oscillations caused by stale states. Experiments on a cycle-accurate simulator show that PSP is robust across port scales, bandwidth-limited paths, and fixed-flow interference. It outperforms join-the-shortest-queue (JSQ) scheduling and Random in loss rate, 99th-percentile buffer occupancy, and scalability, while remaining competitive with Top-k at lower hardware cost. PSP provides an effective balance among performance, stability, and overhead for artificial intelligence data centers.","PeriodicalId":496270,"journal":{"name":"arXiv (Cornell University)","volume":"1 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-08-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://arxiv.org/pdf/2608.08425","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148861346","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Relevance-Aware Rule: Structural Deletion of Irrelevant Conditions in Decision Trees 关联感知规则：决策树中不相关条件的结构性删除

arXiv (Cornell University) Pub Date : 2026-07-15 DOI: 10.48550/arxiv.2607.13874

Jung-Sik Hong, Jeongeon Lee, Min Kyu Sim, Sangheum Hwang

{"title":"Relevance-Aware Rule: Structural Deletion of Irrelevant Conditions in Decision Trees","authors":"Jung-Sik Hong, Jeongeon Lee, Min Kyu Sim, Sangheum Hwang","doi":"10.48550/arxiv.2607.13874","DOIUrl":"https://doi.org/10.48550/arxiv.2607.13874","url":null,"abstract":"Decision trees generate interpretable if--then rules, yet they contain irrelevant conditions (IRCs). These IRCs arise from the structural mechanism of tree splitting and persist even in modern optimal sparse tree induction algorithms. Existing IRC deletion methods overlook this structural mechanism; therefore, they either preserve the original tree too loosely to remain reliable, or too strictly to achieve meaningful simplification. This study provides theoretical foundations for reliable IRC deletion by establishing theorems and propositions related to the underlying IRC mechanism. The key finding is that a binary split shifts class proportions in opposite directions relative to the parent. Specifically, an increase in the class-1 proportion along one branch necessitates an increase in the class-0 proportion along its sibling, thereby generating a C1-link and a C0-link. Based on this structural fact, we propose a structural IRC deletion framework. Relative to each leaf, links that increase the leaf-class proportion are matched, whereas links that increase the proportion of the opposite leaf-class are mismatched. These mismatched links are flagged as structurally suspicious IRC candidates. Rather than deleting them outright, the framework rigorously diagnoses their relevance by assessing prediction reliability. It selectively deletes conditions that are structurally and empirically irrelevant, while strictly protecting those whose deletion would reduce the rule's reliability. Experimental results confirm that the proposed framework achieves substantial rule simplification without sacrificing the reliability of the original tree.","PeriodicalId":496270,"journal":{"name":"arXiv (Cornell University)","volume":"1 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-07-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148861347","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Exploiting Spanning Trees for Directed Acyclicity 利用生成树研究有向非环性

arXiv (Cornell University) Pub Date : 2026-07-08 DOI: 10.48550/arxiv.2607.07705

Sergei Khargeliia, Danil Sagunov

{"title":"Exploiting Spanning Trees for Directed Acyclicity","authors":"Sergei Khargeliia, Danil Sagunov","doi":"10.48550/arxiv.2607.07705","DOIUrl":"https://doi.org/10.48550/arxiv.2607.07705","url":null,"abstract":"We study the weighted case of the textsc{Maximum Acyclic Subgraph (MAS)} problem, where each edge of a given directed graph has a positive weight assigned, and the task is to find a maximum-weight acyclic edge set. The famous and well-studied random ordering lower bound guarantees the existence of an acyclic set that gives at least the half of the total edge weight. The maximum spanning tree (MaxST) guarantee, which is the weight of a maximum-weight acyclic subgraph of the underlying undirected graph of $G$, is another natural lower bound for the weight of an acyclic subgraph. A solution of this weight dominates the random ordering solution on instances where MaxST spans the most of the total edge weight. Our main contribution are two parameterized algorithms that find acyclic subgraphs of total weight larger than the weight of the MaxST of $G$. Both our algorithms find a solution of total weight at least $MaxST(G)+k$, for a given integer $kge 0$, or report that it does not exist, and first of our algorithms runs in time $2^{k^{mathcal{O}(1)}}cdot mathcal{I}^{mathcal{O}(1)}$ and works when all weights are integers; our second algorithm handles rational weights not less than $1$, and its running time is upper-bounded by $n^{k^{mathcal{O}(1)}}cdot mathcal{I}^{mathcal{O}(1)}$. This positive result is rather surprising since solving textsc{MAS} above the random ordering lower bound is classNP-hard in the same rational weights scenario, when $k=1$. Our findings unravel intricate connections between structure of MaxSTs and directed cycles, use perfect graph theorem to tackle rational weights, and raise graph-theoretic questions that are interesting on their own. Of another importance, this is one of the few examples of positive ``above guarantee'' results for a weighted problem on directed graphs, especially for rational weights.","PeriodicalId":496270,"journal":{"name":"arXiv (Cornell University)","volume":"1 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-07-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148861342","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Conical Intersections Enable Ultrafast Molecular Spin Control in a Chromium Complex 锥形交叉使铬络合物的超快分子自旋控制成为可能

arXiv (Cornell University) Pub Date : 2026-06-30 DOI: 10.48550/arXiv.2606.31905

Zihui Liu, Junhua Zhou, T X Chen, Michael Penny, Sara Mosca, Mengyuan Cui, Vandana Tiwari, R. J. Dwayne Miller, Fulu Zheng, Ajay Jha, Hong-Guang Duan

引用次数: 0

An Additive Approximation Scheme for Generating Dyadic Codings for the Outputs of an LLM LLM输出双进编码的一种加性逼近格式

arXiv (Cornell University) Pub Date : 2026-05-07 DOI: 10.48550/arxiv.2605.05837

Daniella Bar-Lev, Farzad Farnoud, Ryan Gabrys

引用次数: 0

Dynamic centrality of headwater sources in river networks: a stochastic approach via ultrametric Laplacians 河网中源头的动态中心性：基于超量程拉普拉斯算子的随机方法

arXiv (Cornell University) Pub Date : 2026-05-05 DOI: 10.48550/arxiv.2605.16328

Ángel Alfredo Morán Ledezma

{"title":"Dynamic centrality of headwater sources in river networks: a stochastic approach via ultrametric Laplacians","authors":"Ángel Alfredo Morán Ledezma","doi":"10.48550/arxiv.2605.16328","DOIUrl":"https://doi.org/10.48550/arxiv.2605.16328","url":null,"abstract":"River networks are hierarchical transport systems in which the timing and position of headwater confluences govern hydrologic response, solute transport, and ecological connectivity. Despite the recognized importance of headwater sources in structuring downstream processes, no mathematically grounded centrality index exists that captures their dynamic role in the transport hierarchy. We apply the dynamic centrality index $C_{mathrm{CTMC}}$ [Morán Ledezma, arXiv:2603.20922], originally introduced in the context of phylogenetic trees, to the problem of headwater centrality in river networks via the dynamic tree representation of [Zaliapin et al., https://doi.org/10.1029/2009JF001281]. Through a topological analysis of the ultrametric structure induced by the dynamic tree, we show that high-centrality headwaters are the tributaries that most efficiently transmit water into the rest of the network, in the sense that their flows merge earliest and most broadly with surrounding sources as transport proceeds downstream. The index admits a fully explicit closed-form expression computable in $O(n)$ time from the tree structure alone, without simulation. Comparing $C_{mathrm{CTMC}}$ rankings against the number of downstream junctions reached during transport, a direct measure of hydrological influence, on a dataset of 49 natural river basins across the United States, we find that top-ranked headwaters consistently reach a disproportionately large number of junctions across all transport times. This indicates that high-centrality headwaters are not merely early contributors but consistently influential throughout the entire transport process. These results suggest that ultrametric spectral analysis provides an interpretable and scalable framework for identifying hydrologically influential headwaters, with potential applications in ecological monitoring and watershed management.","PeriodicalId":496270,"journal":{"name":"arXiv (Cornell University)","volume":"1 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-05-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148305652","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

SprayCheck: Finding Gray Failures in Adaptive Routing Networks SprayCheck：查找自适应路由网络中的灰色故障

arXiv (Cornell University) Pub Date : 2026-05-05 DOI: 10.48550/arxiv.2605.03702

Jakob Krebs, Daniel Amir, Shir Landau Feibish, Mark Silberstein

{"title":"SprayCheck: Finding Gray Failures in Adaptive Routing Networks","authors":"Jakob Krebs, Daniel Amir, Shir Landau Feibish, Mark Silberstein","doi":"10.48550/arxiv.2605.03702","DOIUrl":"https://doi.org/10.48550/arxiv.2605.03702","url":null,"abstract":"Distributed machine learning (ML) training has become a dominant workload in modern data center networks, operating at massive scale with clusters comprising tens to hundreds of thousands of GPUs. The scale of these networks makes failures, and particularly gray failures, inevitable. Gray failures can significantly degrade both network and application performance, yet they are notoriously difficult to detect, localize, and debug. To meet the performance demands of ML workloads, adaptive routing is widely deployed to maximize network utilization by dynamically spreading traffic across many paths. While adaptive routing increases network utilization, it also greatly intensifies the effect of gray failures. Prior work has either dismissed gray failures as negligible or proposed detection mechanisms that fail to scale, rendering these approaches increasingly impractical for large-scale clusters. We present SprayCheck, a passive gray failure detection system that leverages the statistical properties of adaptive routing and network load balancing. By combining these properties with flow-level information, SprayCheck can identify failures before they significantly impact application performance, enabling preemptive rerouting and improving overall performance. Importantly, this is achieved through passive observation of traffic spraying, without introducing additional load on the network. We evaluate SprayCheck and show that it can detect and localize a single-link packet-drop-rate $1.5%$ within a single iteration and as little as $0.5%$ within 5 training iterations of Llama-3 70B in a 64 spine topology.","PeriodicalId":496270,"journal":{"name":"arXiv (Cornell University)","volume":"1 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-05-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148672256","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Low-Rank Approximation by Randomly Pivoted LU 随机中心LU的低秩逼近

arXiv (Cornell University) Pub Date : 2026-01-29 DOI: 10.48550/arxiv.2601.22344

Marc Aurèle Gilles, Heather Wilber

引用次数: 0