• 学术搜索
  • 科研智能体
    • Research Labs
    • AI 阅读
    • AI 文库
    • 深度研究
    • 学者亮点
  • 学术资源
    • AI2000
    • 期刊/会议
    • 学者库
    • 学术API
      MCP&Skill
    • 溯源树
    • 数据集
  • 知识沉淀
    • 学术空间
订阅小程序
旧版功能
aminer vip
开通会员低至0.73元/天
一次搞定AI科研
立即登录
  • English
  • 联系方式
    网

    网络器械公司

    NetApp
    企业
    200论文总数
    9,656引用总数

    NetApp(纳斯达克股票代码:NTAP),不断以创新的理念和领先的技术引领存储行业的发展,是向数据密集型企业提供统一存储解决方案的居世界最前列的公司,其 Data ONTAP是全球首屈一指的存储操作系统。自 1992 年创立以来,NetApp一直在率先提供一流的技术、产品和合作关系,提升IT效率和灵活性的储存和数据管理解决方案,不断推动存储业的发展。NetApp 的存储解决方案涵盖了专业化的硬件、软件和服务,为开放网络环境提供了无缝的存储管理。

    论文量&引用量时间轴

    机构学者

    论文(200)

    年份
    起
    –
    止
    排序
    1Your Embedding Model is SMARTer Than You Think
    Jianrui Zhang,Hyun Jung Lee, Sukanta Ganguly, Tae-Eui Kam,Donghyun Kim,Yong Jae Lee

    Multimodal retrieval relies heavily on single-vector retrievers, which compress rich, sequential token sequences into one single global representation. While efficient, they discard fine-grained, local evidence critical for dense retrieval tasks. Multi-vector approaches were introduced as a solution, but they strictly require training and many ignore the necessity of a globally summarizing representation. To address this, we introduce SMART, a framework that unlocks the latent multi-vector capabilities of standard single-vector models. We first demonstrate that standard contrastive training on the pooled embedding implicitly shapes the retrieval geometry of preceding hidden states via gradient flow. By applying direct late-interaction over these frozen hidden states during inference, SMART acts as a plug-and-play upgrade that consistently improves performance across diverse modalities, pushing even the state-of-the-art models to new heights on MMEB-V2. We further reveal SMART's superior performance as simple lightweight post-training not only saves time and compute, but also brings forth further improvement on Visual Document retrieval, allowing a single-vector model to outperform SoTA multi-vector counterparts. Ultimately, SMART offers both a highly efficient inference enhancement and a powerful finetuning technique for multimodal retrieval.

    2026NeurIPS 2026(2026)引用:2
    引用
    AI阅读
    加入学术空间
    2Reasoning-Augmented Representations for Multimodal Retrieval
    Jianrui Zhang,Anirudh Sundara Rajan, Brandon Han,Soochahn Lee, Sukanta Ganguly,Yong Jae Lee

    Universal Multimodal Retrieval (UMR) seeks any-to-any search across text and vision, yet modern embedding models remain brittle when queries require latent reasoning (e.g., resolving underspecified references or matching compositional constraints). We argue this brittleness is often data-induced: when images carry "silent" evidence and queries leave key semantics implicit, a single embedding pass must both reason and compress, encouraging spurious feature matching. We propose a data-centric framework that decouples these roles by externalizing reasoning before retrieval. Using a strong Vision–Language Model, we make implicit semantics explicit by densely captioning visual evidence in corpus entries, resolving ambiguous multimodal references in queries, and rewriting verbose instructions into concise retrieval constraints. Inference-time enhancement alone is insufficient; the retriever must be trained on these semantically dense representations to avoid distribution shift and fully exploit the added signal. Across M-BEIR, our reasoning-augmented training method yields consistent gains over strong baselines, with ablations showing that corpus enhancement chiefly benefits knowledge-intensive queries while query enhancement is critical for compositional modification requests. We publicly release our code at https://github.com/AugmentedRetrieval/ReasoningAugmentedRetrieval.

    2026CoRR(2026)引用:2
    引用
    AI阅读
    加入学术空间
    3Leafblower: a Leakage Attack Against Tee-Based Encrypted Databases
    Zachary Espiritu,Seny Kamara,Tarik Moataz, Valentin Ogier

    Trusted execution environments (TEEs) have emerged as a common solution for database systems to provide encryption in use. Several encrypted databases (EDBs) have been deployed within TEEs using library operating system toolchains that transparently allow existing applications to run within TEEs without modification. This “lift-and-shift” paradigm greatly simplifies the design of EDBs but leaves open questions about the security of the resulting system. In this work, we propose a new leakage attack against TEEbased EDBs which use $\mathbf{B}^{+}$-trees in the multi-snapshot external memory model, a weaker adversary which only observes snapshots of the encrypted database index files after each operation. We show how to approximately order insertions by their inserted value by exploiting the “structural leakage” of the on-disk index format. Then, we leverage auxiliary information to recover the approximate plaintext values of insert operations with significant advantage over a naive adversary that makes guesses based on equivalent auxiliary information. Under optimal conditions-when the auxiliary is accurate and the domain is small-we achieve up to 96% exact recovery in experiments on real-world datasets which increases to 100% when scoped to later operations in the transcript. Our attack requires no injections and no information about read operations. While our work is primarily motivated by TEE-based encrypted databases, we demonstrate that our attack generalizes to other kinds of page-level encryption systems including encrypted storage engines and disaggregated database systems.

    20262026 IEEE Symposium on Security and Privacy (SP)(2026)引用:1
    引用
    AI阅读
    加入学术空间
    4Mitigating False Collisions in Partition-Based Concurrency Control
    Matthew Curtis-Maury, Sushrut Bhowmik, Grey Files, Jian Hu, Bryan Prosser

    Modern storage systems must scale request processing across many cores, making efficient concurrency control essential. Data partitioning routes operations to execution contexts that own portions of system state, reducing synchronization overhead but potentially introducing false collisions: unnecessary serialization imposed by partition boundaries rather than true data dependencies. False collisions arise when operations span partitions (e.g., large or misaligned I/Os) or when static provisioning mismatches demand, reducing parallelism and increasing queueing delay. We present two complementary mechanisms to mitigate false collisions in partitioned execution frameworks. Relaxed Data Partitioning (RDP) executes common partition-spanning I/Os in fine-grained partitions while synchronizing only the additional accessed regions via targeted, partition-aware locking. Elastic Data Partitioning (EDP) dynamically adjusts the number of active execution partitions through auxiliary execution contexts, increasing parallelism without changing ownership or requiring online repartitioning. We instantiate RDP and EDP in WAFL's hierarchical partitioning framework and evaluate them on multiple platforms and representative request streams. RDP improves throughput by up to 105%, EDP by up to 82%, and together by up to 145%, while imposing negligible overhead.

    2026PROCEEDINGS OF THE 19TH ACM INTERNATIONAL SYSTEMS AND STORAGE CONFERENCE, SYSTOR 2026(2026)
    引用
    AI阅读
    加入学术空间
    5Streamlining DevOps Pipelines with AI-Augmented Feature Flagging for Microservices Architectures
    Ravi Babu Dasari

    The complexity of microservices-based SaaS applications, coupled with the demands of rapid feature development, poses significant challenges for DevOps pipelines, particularly in managing feature rollouts for large teams. This article proposes an AI-augmented feature flagging system that enhances deployment efficiency by predicting feature stability, optimizing rollout strategies, and automating testing configurations. The system relies on machine learning models, the AWS Bedrock API, and the OpenAI API to identify the changes in the code and runtime metrics and use these to guide flag management, which is combined with Kubernetes and CI/CD, such as Jenkins. A React-based dashboard gives the product managers, QA engineers, and developers real-time insight into the rollout progress. This is based on real-world simulations based upon large-scale deployments of microservices and shows a significant decrease in deployment errors and quicker release cycles. The architectural framework integrates predictive intelligence and automated optimization in order to overcome important issues of development teams that deal with complex distributed systems. Machine learning potential will be integrated into the existing DevOps practices to optimize human decision-making during the deployment lifecycle. The numerous benefits of an organization adopting AI-enhanced feature flagging are an increase in deployment reliability, a shortened release schedule, lowered infrastructure expenses, and the development of strong teamwork. The benefits to the environment could also be seen regarding optimal use of resources and reduced energy use in cloud computing infrastructure. The economic returns are in the form of cost savings, productivity, as well as reduced time-to-market in revenue-generating features. The system makes feature flag management more of a proactive risk management process rather than a problem-solving approach. Gradual implementation plans allow building organizational capability stepwise and prove its value at every step. These governance regimes guarantee the right human control and have real control of important issues of deployment. The modular architectural design will be compatible with the existing DevOps toolchains and allow gradual improvement as the organizational capabilities advance.

    2026Journal of Computer Science and Technology Studies(2026)
    引用
    AI阅读
    加入学术空间
    立即登录,查看全部 200 篇论文

    合作机构(100)

    Helmholtz Center for Information Security,Helmholtz Association of German Research Centres合作论文 7
    威斯康星大学麦迪逊分校合作论文 5
    德国亥姆霍兹研究中心协会合作论文 5
    伊利诺伊大学香槟分校合作论文 4
    西安交通大学合作论文 4
    波士顿大学合作论文 4
    加利福尼亚大学圣地亚哥分校合作论文 4
    Ministerium der Finanzen des Landes Sachsen-Anhalt合作论文 3
    微软合作论文 3
    威廉与玛丽学院合作论文 3

    机构统计