Open-domain textual question answering (QA), which aims to answer questions from large data sources like Wikipedia or the web, has gained wide attention in recent years. Recent advancements open-domain QA are mainly due significant developments of deep learning techniques, especially machine reading comprehension and neural-network-based information retrieval, allows models continuously refresh state-of-the-art performances. However, a comprehensive review existing approaches trends is...

Most existing author disambiguation work relies heavily on feature engineering or cannot use multiple paper relationships. In this work, we propose a network-embedding based method for disambiguation. For each ambiguous name, construct networks among papers sharing an and connect with relationships (e.g., co-authoring paper). We focus maximizing the gap between positive edges negative edges, graph coarsening technique to learn global information. Further, design clustering algorithm which...

Traffic classification associates packet streams with known application labels, which is vital for network security and management. With the rise of NAT, port dynamics, encrypted traffic, it increasingly challenging to obtain unified traffic features accurate classification. Many state-of-the-art classifiers automatically extract from stream based on deep learning models such as convolution networks. Unfortunately, compositional causal relationships between packets are not well extracted in...

Network traffic classification is important for network security and management. State-of-the-art classifiers use deep learning techniques to automatically extract feature vectors from the traffic, which however lose context of communication sessions encapsulated text semantics. In this paper, we present a Multi-Modal Classification method named MTCM systematically exploit task. We build an adaptive context-aware extraction framework over varying-length dynamic packet sequences, based on...

Distributed data-parallel training (DDP) is prevalent in large-scale deep learning. To increase the throughput and scalability, high-performance collective communication methods such as AllReduce have recently proliferated for DDP use. However, these approaches require long periods with increasing model sizes. Collective transmits many sparse gradient values that can be efficiently compressed to reduce required time. State-of-the-art compression do not provide mergeable lack convergence...

Network monitoring is vital in modern clouds and data center networks that need diverse traffic statistics ranging from flow size distributions to heavy hitters. To cope with increasing network rates massive volumes, sketch based approximate measurement has been extensively studied trade the accuracy for memory computation cost, which unfortunately, sensitive hash collisions.This paper presents a clustering-preserving method be resilient collisions. We provide an equivalence analysis of...

The network distance service obtains the latency among large-scale nodes. With increasing numbers of participating nodes, has to balance accuracy and scalability. network-coordinate methods scale well by embedding pairwise into a low-dimensional coordinate system. prediction errors are iteratively optimized adjusting coordinates with respect neighbors. Unfortunately, optimization process is vulnerable inaccurate coordinates, leading destabilized positions. In this paper, we propose RMF,...

Network traffic classification is crucial for network security and management one of the most important tasks. Current state-of-the-art classifiers are based on deep learning models to automatically extract features from packet streams. Unfortunately, current approaches fail effectively combine structural information packets with content packets, resulting in limited accuracy. In this paper, we propose a graph neural model classification, which can well perceive interaction feature traffic....

Distributed online social networks (DOSN) have emerged recently. Nevertheless, recommending friends in the distributed has not been exploited fully. We propose BCE (Bloom Filter based Common-Friend Estimation), a scalable and privacy-preserving common-friend estimation scheme that estimates set of common without need cryptography techniques. First, denotes each user using identifiers created by Peer-to-Peer underlay are robust against dictionary attacks. Second, uses Bloom filter to...

Many persons share with the same name. Distinguishing different name is important but challenging. Albeit much work has been proposed for author disambiguation, most of them do not adequately consider heterogeneous relationships among authors and papers. In our work, ambiguous names their related information, such as papers, conferences, titles, abstracts, etc., are constructed into a network which consists edge types. To fully incorporate all information network, we use Generative...

Withdrawn.

It is important to disambiguate names among persons in many scenarios. In this work, we propose an unsupervised method Diting and a semi-supervised Diting++ for author disambiguation. Diting, learn low-dimensional vector represent each paper networks, which are formed by connecting papers with multiple types of relationship (such as co-author). During representation learning, focus on maximizing the gap between positive edges negative edges. Further, clustering algorithm associates their...

The communication among agents is important for Multi-Agent Reinforcement Learning (MARL). In this work, we propose GraphComm, a method makes use of the relation-ships MARL communication. GraphComm takes explicit relations (e.g., agent types), which can be provided through some knowledge background, into account to better model relationships agents. Besides relations, considers implicit are formed by interactions. Graph Neural Networks (GNNs) relational information, and GNNs assist learning...

Graph neural networks (GNNs) have been successfully applied to many important application domains on graph data. As graphs become increasingly large, existing GNN training frameworks typically use mini-batch sampling during feature aggregation lower resource burdens, which unfortunately suffer from long memory accessing latency and inefficient data transfer of vertex features CPU GPU. This paper proposes 2PGraph, a system that addresses these limitations supports fast efficient single-GPU...

User-facing services deployed in data centers must respond quickly to user actions. The measurement of network latencies is paramount importance. Recently, a new family compact structures has been proposed estimate one-way latencies. In order achieve scalability, these methods rely on timestamp aggregation. Unfortunately, this approach suffers from serious accuracy problems the presence packet loss and reordering, given that single lost or out-of-order may invalidate huge number aggregated...

Predicting network latencies between Internet hosts can efficiently support large-scale applications, e.g., file sharing service and the overlay construction. Several study use hyperbolic space to model dense-core many-tendril structure. However, existing based embedding approaches are not designed for accurate latency estimation in distributed context. We present HyperSpring, which estimates by modelling a mass spring system similar with Vivaldi. HyperSpring adopts coordinate initialization...

User generated video systems like YouTube and Twitch.tv have been a major internet phenomenon. They attracted vast user base with their many varied contents provided by users, series of social features tailored for online viewing. In hoping building more lively community encouraging the content creators to share more, recently such introduced crowdsourcing mechanisms wherein get tangible rewards through donations. donation is very special form relationships. It influences engagement in...

Online cloud services need to fulfill clients' requests scalably and fast, otherwise, users' experience providers' revenue could be severely degraded.State-of-the-art are increasingly deployed as a distributed service mesh.Service communication is frequent in the mesh.Unfortunately, problematic events may occur between any pair of nodes mesh, therefore, it vital maximize network visibility for efficient troubleshooting application optimization.A state-of-the-art approach model pairwise RTTs...