检索结果-内蒙古大学图书馆

RADMM: RECURRENT ADAPTIVE MIXTURE MODEL WITH APPLICATIONS TO DOMAIN ROBUST LANGUAGE MODELING

学校读者我要写书评

暂无评论

RADMM: RECURRENT ADAPTIVE MIXTURE MODEL WITH APPLICATIONS TO...

IEEE International Conference on Acoustics, Speech and Signal Processing

作者： Kazuki Irie Shankar Kumar Michael Nirschl Hank Liao Human Language Technology and Pattern Recognition Group Computer Science Department RWTH Aachen University D-52056 Aachen Germany Google Inc. New York NY 10011 USA

ISBN: (纸本)9781538646595

We present a new architecture and a training strategy for an adaptive mixture of experts with applications to domain robust language modeling. The proposed model is designed to benefit from the scenario where the training data are available in diverse domains as is the case for YouTube speech recognition. The two core components of our model are an ensemble of parallel long short-term memory (LSTM) expert layers for each domain and another LSTM based network which generates state dependent mixture weights for combining expert LSTM states by linear interpolation. The resulting model is a recurrent adaptive mixture model (RADMM) of domain experts. We train our model on 4.4B words from YouTube speech recognition data. We report results on the YouTube speech recognition test set. Compared with a background LSTM model, we obtain up to 12% relative improvement in perplexity and an improvement in word error rate from 12.3% to 12.1% while using a lattice rescoring with strong pruning.

关键词： language modeling neural networks speech recognition mixture of experts domain adaptation modelling languages Speech recognition Neural network Professional personnel YouTube Mixture models training policy new buildings

Managing recurrent virtual network updates in multi-tenant datacenters: A system perspective

学校读者我要写书评

暂无评论

arXiv 2019年

作者： Liu, Zhuotao Cao, Yuan Zhang, Xuewu Zhu, Changping Zhang, Fan Electrical and Computer Engineering Department University of Illinois at Urbana-Champaign UrbanaIL61801 United States Google Inc. Mountain ViewCA94043 United States College of Internet of Things Engineering Hohai University Changzhou213022 China Key Laboratory of Unmanned Aerial Vehicle Development and Data Application of Anhui Higher Education Institutes Wanjiang Institute of Technology Anhui243031 China College of Information Science and Electronic Engineering Zhejiang University ZheJiang310027 China

With the advent of software-defined networking, network configuration through programmable interfaces becomes practical, leading to various on-demand opportunities for network routing update in multi-tenant datacenters, where tenants have diverse requirements on network routings such as short latency, low path inflation, large bandwidth, high reliability, etc. Conventional solutions that rely on topology search coupled with an objective function-https://***/project/5beb742041ab9c0e3caec84f to find desired routings have at least two shortcomings: (i) they run into scalability issues when handling consistent and frequent routing updates and (ii) they restrict the flexibility and capability to satisfy various routing requirements. To address these issues, this paper proposes a novel search and optimization decoupled design, which not only saves considerable topology search costs via search result reuse, but also avoids possible sub-optimality in greedy routing search algorithms by making decisions based on the global view of all possible routings. We implement a prototype of our proposed system, OpReduce, and perform extensive evaluations to validate its design goals. Copyright © 2019, The Authors. All rights reserved.

关键词： Network routing

Online resource allocation under partially predictable demand

学校读者我要写书评

暂无评论

arXiv 2018年

作者： Hwang, Dawsen Jaillet, Patrick Manshadi, Vahideh Google ChicagoIL60607 Department of Electrical Engineering and Computer Science Laboratory for Information and Decision Systems Operations Research Center Massachusetts Institute of Technology CambridgeMA02139 Yale School of Management New HavenCT06511

For online resource allocation problems, we propose a new demand arrival model where the sequence of arrivals contains both an adversarial component and a stochastic one. Our model requires no demand forecasting;however, due to the presence of the stochastic component, we can partially predict future demand as the sequence of arrivals unfolds. Under the proposed model, we study the problem of the online allocation of a single resource to two types of customers, and design online algorithms that outperform existing ones. Our algorithms are adjustable to the relative size of the stochastic component, and our analysis reveals that as the portion of the stochastic component grows, the loss due to making online decisions decreases. This highlights the value of (even partial) predictability in online resource allocation. We impose no conditions on how the resource capacity scales with the maximum number of customers. However, we show that using an adaptive algorithm-which makes online decisions based on observed data-is particularly beneficial when capacity scales linearly with the number of customers. Our work serves as a first step in bridging the longstanding gap between the two well-studied approaches to the design and analysis of online algorithms based on (1) adversarial models and (2) stochastic ones. Using novel algorithm design, we demonstrate that even if the arrival sequence contains an adversarial component, we can take advantage of the limited information that the data reveals to improve allocation decisions. We also study the classical secretary problem under our proposed arrival model, and we show that randomizing over multiple stopping rules may increase the probability of success. Copyright © 2018, The Authors. All rights reserved.

关键词： Resource allocation

Optimization Framework for Minimizing Rule Update Latency in SDN Switches

学校读者我要写书评

暂无评论

ZTE Communications 2018年第4期16卷 15-29页

作者： CHEN Yan WEN Xitao LENG Xue YANG Bo Li Erran Li ZHENG Peng HU Chengchen College of Computer Science and Technology Zhejiang UniversityHangzhou 310027China Department of Electrical Engineering and Computer Science Northwestern UniversityEvanstonIL 60208USA Google Inc. Mountain ViewCA 94043USA Microsoft Shanghai 200000China Uber Technologies Inc. San FranciscoCA 94103USA School of Electronic and Information Engineering Xi’an Jiaotong UniversityXi’an 710049China

Benefited from the design of separating control plane and data plane,software defined networking(SDN)is widely concerned and *** quick response capability to network events with changes in network policies enables more dynamic management of data center *** the SDN controller architecture is increasingly optimized for swift policy updates,the data plane,especially the prevailing ternary content-addressable memory(TCAM)based flow tables on physical SDN switches,remains unoptimized for fast rule updates,and is gradually becoming the primary bottleneck along the policy update *** this paper,we present RuleTris,the first SDN update optimization framework that minimizes rule update latency for TCAM-based *** employs the dependency graph(DAG)as the key abstraction to minimize the update *** efficiently obtains the DAGs with novel dependency preserving algorithms that incrementally build rule dependency along with the compilation ***,in the guidance of the DAG,RuleTris calculates the TCAM update schedules that minimize TCAM entry moves,which are themain cause of TCAM update *** evaluation,RuleTris achieves a median of<12 ms and 90-percentile of<15ms the end-to-end perrule update latency on our hardware prototype,outperforming the state-of-the-art composition compiler CoVisor by～20 times.

关键词： SDN SDN-based cloud network management access control unauthorized attack

Navigating weighted regions with scattered skinny tetrahedra

学校读者我要写书评

暂无评论

International Journal of Computational Geometry and Applications 2017年第1-2期27卷 13-32页

作者： Cheng, Siu-Wing Chiu, Man-Kwun Jin, Jiongxin Vigneron, Antoine Department of Computer Science and Engineering HKUST Hong Kong National Institute of Informatics (NII) Tokyo Japan JST ERATO Kawarabayashi Large Graph Project Japan Google Inc. United States School of Electrical and Computer Engineering Ulsan National Institute of Science and Technology Ulsan South Korea

We propose an algorithm for finding a (1 + ϵ)-approximate shortest path through a weighted 3D simplicial complex τ. The weights are integers from the range [1,W] and the vertices have integral coordinates. Let N be the largest vertex coordinate magnitude, and let n be the number of tetrahedra in τ. Let ρ be some arbitrary constant. Let Κ be the size of the largest connected component of tetrahedra whose aspect ratios exceed Κ. There exists a constant C dependent on Κ but independent of such that if Κ ≤ 1 Cloglog n + O(1), the running time of our algorithm is polynomial in n, 1/ and log(NW). If Κ = O(1), the running time reduces to O(nϵ-O(1)(log(NW))O(1)). © 2017 World Scientific Publishing Company.

关键词： Approximation algorithm Shortest path Weighted region

Readiness of Quantum Optimization Machines for Industrial Applications

学校读者我要写书评

暂无评论

Physical Review Applied 2019年第1期12卷 014004-014004页

作者： Alejandro Perdomo-Ortiz Alexander Feldman Asier Ozaeta Sergei V. Isakov Zheng Zhu Bryan O’Gorman Helmut G. Katzgraber Alexander Diedrich Hartmut Neven Johan de Kleer Brad Lackey Rupak Biswas Quantum Artificial Intelligence Lab. NASA Ames Research Center Moffett Field California 94035 USA USRA Research Institute for Advanced Computer Science (RIACS) Mountain View California 94043 USA Zapata Computing Inc. 439 University Avenue Office 535 Toronto Ontario M5G 1Y8 Canada Department of Computer Science University College London WC1E 6BT London United Kingdom Palo Alto Research Center 3333 Coyote Hill Road Palo Alto California 94304 USA QC Ware Corp. 125 University Avenue Suite 260 Palo Alto California 94301 USA Google Inc. 8002 Zurich Switzerland Department of Physics and Astronomy Texas A&M University College Station Texas 77843-4242 USA Berkeley Center for Quantum Information and Computation Berkeley California 94720 USA Department of Chemistry University of California Berkeley California 94720 USA 1QB Information Technologies (1QBit) Vancouver British Columbia V6B 4W4 Canada Santa Fe Institute 1399 Hyde Park Road Santa Fe New Mexico 87501 USA Fraunhofer IOSB-INA Lemgo Germany Google Inc. Venice California 90291 USA Joint Center for Quantum Information and Computer Science University of Maryland College Park Maryland 20742 USA Departments of Computer Science and Mathematics University of Maryland College Park Maryland 20742 USA Mathematics Research Group National Security Agency Ft. George G. Meade Maryland 20755 USA Exploration Technology Directorate NASA Ames Research Center Moffett Field California 94035 USA

There have been multiple attempts to demonstrate that quantum annealing and, in particular, quantum annealing on quantum-annealing machines, has the potential to outperform current classical optimization algorithms implemented on CMOS technologies. The benchmarking of these devices has been controversial. Initially, random spin-glass problems were used, however, these were quickly shown to be not well suited to detect any quantum speedup. Subsequently, benchmarking shifted to carefully crafted synthetic problems designed to highlight the quantum nature of the hardware while (often) ensuring that classical optimization techniques do not perform well on them. Even worse, to date a true sign of improved scaling with the number of problem variables remains elusive when compared to classical optimization techniques. Here, we analyze the readiness of quantum-annealing machines for real-world application problems. These are typically not random and have an underlying structure that is hard to capture in synthetic benchmarks, thus posing unexpected challenges for optimization techniques, both classical and quantum alike. We present a comprehensive computational scaling analysis of fault diagnosis in digital circuits, considering architectures beyond D-Wave quantum annealers. We find that the instances generated from real data in multiplier circuits are harder than other representative random spin-glass benchmarks with a comparable number of variables. Although our results show that transverse-field quantum annealing is outperformed by state-of-the-art classical optimization algorithms, these benchmark instances are hard and small in the size of the input, therefore representing the first industrial application ideally suited for testing near-term quantum annealers and other quantum algorithmic strategies for optimization problems.

关键词： Adiabatic quantum optimization Quantum algorithms Quantum benchmarking Quantum information architectures & platforms Quantum Monte Carlo

Multi-view network embedding via graph factorization clustering and co-regularized multi-view agreement

学校读者我要写书评

暂无评论

arXiv 2018年

作者： Sun, Yiwei Bui, Ngot Hsieh, Tsung-Yu Honavar, Vasant Artificial Intelligence Research Laboratory Pennsylvania State University United States College of Information Sciences and Technology Pennsylvania State University United States Department of Computer Science and Engineering Pennsylvania State University United States Google LLC United States

Real-world social networks and digital platforms are comprised of individuals (nodes) that are linked to other individuals or entities through multiple types of relationships (links). Sub-networks of such a network based on each type of link correspond to distinct views of the underlying network. In real-world applications, each node is typically linked to only a small subset of other nodes. Hence, practical approaches to problems such as node labeling have to cope with the resulting sparse networks. While low-dimensional network embeddings offer a promising approach to this problem, most of the current network embedding methods focus primarily on single view networks. We introduce a novel multi-view network embedding (MVNE) algorithm for constructing low-dimensional node embeddings from multi-view networks. MVNE adapts and extends an approach to single view network embedding (SVNE) using graph factorization clustering (GFC) to the multi-view setting using an objective function that maximizes the agreement between views based on both the local and global structure of the underlying multi-view graph. Our experiments with several benchmark real-world single view networks show that GFC-based SVNE yields network embeddings that are competitive with or superior to those produced by the state-of-the-art single view network embedding methods when the embeddings are used for labeling unlabeled nodes in the networks. Our experiments with several multi-view networks show that MVNE substantially outperforms the single view methods on integrated view and the state-of-the-art multi-view methods. We further show that even when the goal is to predict labels of nodes within a single target view, MVNE outperforms its single-view counterpart suggesting that the MVNE is able to extract the information that is useful for labeling nodes in the target view from the all of the views. Copyright © 2018, The Authors. All rights reserved.

关键词： Network embeddings

Disentangling dynamics and content for control and planning

学校读者我要写书评

暂无评论

arXiv 2017年

作者： Banijamali, Ershad Khajenezhad, Ahmad Ghodsi, Ali Ghavamzadeh, Mohammad School of Computer Science University of Waterloo Sharif University of Technology Department of Statistics and Actuarial Science University of Waterloo DeepMind

In this paper, We study the problem of learning a controllable representation for high-dimensional observations of dynamical systems. Specifically, we consider a situation where there are multiple sets of observations of dynamical systems with identical underlying dynamics. Only one of these sets has information about the effect of actions on the observation and the rest are just some random observations of the system. Our goal is to utilize the information in that one set and find a representation for the other sets that can be used for planning and ling-term prediction. Copyright © 2017, The Authors. All rights reserved.

关键词： Dynamical systems

A community-powered search of machine learning strategy space to find NMR property prediction models

学校读者我要写书评

暂无评论

arXiv 2020年

作者： Bratholm, Lars A. Gerrard, Will Anderson, Brandon Bai, Shaojie Choi, Sunghwan Dang, Lam Hanchar, Pavel Howard, Addison Huard, Guillaume Kim, Sanghoon Kolter, Zico Kondor, Risi Kornbluth, Mordechai Lee, Youhan Lee, Youngsoo Mailoa, Jonathan P. Nguyen, Thanh Tu Popovic, Milos Rakocevic, Goran Reade, Walter Song, Wonho Stojanovic, Luka Thiede, Erik H. Tijanic, Nebojsa Torrubia, Andres Willmott, Devin Butts, Craig P. Glowacki, David R. Participants, Kaggle School of Chemistry University of Bristol Cantock’s Close BristolBS8 1TS United Kingdom School of Mathematics University of Bristol Fry Building Woodland Road BristolBS1 1UG United Kingdom Dept. of Computer Science University of Bristol Merchant Venturer’s Building BristolBS8 1UB United Kingdom Intangible Realities Laboratory University of Bristol Cantock’s Close BristolBS8 1TS United Kingdom Bosch Center for Artificial Intelligence PittsburghPA15222 United States Carnegie Mellon University PittsburghPA15213 United States National Institute of Supercomputing and Network Korea Institute of Science and Technology Information 245 Daehak-ro Yuseong-gu Daejeon34141 Korea Republic of BNP Paribas Cardif Fyusion Inc. Kaggle Google Inc. Mountain View United States Ebay Korea 34F Gangnam Finance Center 152 Teheran Ro Gangnam Gu Seoul Korea Republic of Department of Computer Science University of Chicago 5730 S Ellis Ave ChicagoIL60637 United States Department of Statistics University of Chicago 5747 S Ellis Ave ChicagoIL60637 United States Center for Computational Mathematics Flatiron Institute 162 5th Ave. New YorkNY10010 United States Bosch Research and Technology Center CambridgeMA02139 United States Department of Chemical and Biomolecular Engineering Korea Advanced Institute of Science and Technology 291 Daehak-ro Yuseong-gu Daejeon34141 Korea Republic of 111 Daedeok-daero 989beon-gil Yuseong-gu Daejeon Korea Republic of MINDS AND COMPANY 2621 Nambusunhwan-ro Gangnam-gu Seoul06267 Korea Republic of Totient Inc Sindjeliceva 9 Belgrade11000 Serbia KAIST Web Security & Privacy Lab 291 Daehak-ro Yuseong-guDaejeon34141 Korea Republic of *** Alicante Spain

The rise of machine learning (ML) has created an explosion in the potential strategies for using data to make scientific predictions. For physical scientists wishing to apply ML strategies to a particular domain, it can be difficult to assess in advance what strategy to adopt within a vast space of possibilities. Here we outline the results of an online community-powered effort to swarm search the space of ML strategies and develop algorithms for predicting atomic-pairwise nuclear magnetic resonance (NMR) properties in molecules. Using an open-source dataset, we worked with Kaggle to design and host a 3-month competition which received 47,800 ML model predictions from 2,700 teams in 84 countries. Within 3 weeks, the Kaggle community produced models with comparable accuracy to our best previously published ‘in-house’ efforts. A meta-ensemble model constructed as a linear combination of the top predictions has a prediction accuracy which exceeds that of any individual model, 7-19x better than our previous state-of-the-art. The results highlight the potential of transformer architectures for predicting quantum mechanical (QM) molecular properties. Copyright © 2020, The Authors. All rights reserved.

关键词： Nuclear magnetic resonance