检索结果-内蒙古大学图书馆

Towards Dynamic and Safe Configuration Tuning for Cloud databases

学校读者我要写书评

暂无评论

arXiv 2022年

作者： Zhang, Xinyi Wu, Hong Li, Yang Tan, Jian Li, Feifei Cui, Bin School of CS Key Laboratory of High Confidence Software Technologies Peking University China Database and Storage Laboratory Damo Academy Alibaba Group China Center for Data Science Peking University National Engineering Laboratory for Big Data Analysis and Applications China China

Configuration knobs of database systems are essential to achieve high throughput and low latency. Recently, automatic tuning systems using machine learning methods (ML) have shown to find better configurations compared to experienced database administrators (DBAs). However, there are still gaps to apply the existing systems in production environments, especially in the cloud. First, they conduct tuning for a given workload within a limited time window and ignore the dynamicity of workloads and data. Second, they rely on a copied instance and do not consider the availability of the database when sampling configurations, making the tuning expensive, delayed, and unsafe. To fill these gaps, we propose OnlineTune, which tunes the online databases safely in changing cloud environments. To accommodate the dynamicity, OnlineTune embeds the environmental factors as context feature and adopts contextual Bayesian Optimization with context space partition to optimize the database adaptively and scalably. To pursue safety during tuning, we leverage the black-box and the white-box knowledge to evaluate the safety of configurations and propose a safe exploration strategy via subspace adaptation. We conduct evaluations on dynamic workloads from benchmarks and real-world workloads. Compared with the state-of-the-art methods, OnlineTune achieves 14.4%∼165.3% improvement on cumulative performance while reducing 91.0%∼99.5% unsafe configuration recommendations. Copyright © 2022, The Authors. All rights reserved.

关键词： database systems

A regularized approach to sparse optimal policy in reinforcement learning

学校读者我要写书评

暂无评论

arXiv 2019年

作者： Li, Xiang Yang, Wenhao Zhang, Zhihua School of Mathematical Sciences Peking University Beijing China Center for Data Science Peking University Beijing China National Engineering Lab for Big Data Analysis and Applications School of Mathematical Sciences Peking University Beijing China

We propose and study a general framework for regularized Markov decision processes (MDPs) where the goal is to find an optimal policy that maximizes the expected discounted total reward plus a policy regularization term. The extant entropy-regularized MDPs can be cast into our framework. Moreover, under our framework, many regularization terms can bring multi-modality and sparsity, which are potentially useful in reinforcement learning. In particular, we present sufficient and necessary conditions that induce a sparse optimal policy. We also conduct a full mathematical analysis of the proposed regularized MDPs, including the optimality condition, performance error, and sparseness control. We provide a generic method to devise regularization forms and propose off-policy actor critic algorithms in complex environment settings. We empirically analyze the numerical properties of optimal policies and compare the performance of different sparse regularization forms in discrete and continuous environments. Copyright © 2019, The Authors. All rights reserved.

关键词： Markov processes

Battery Valuation and Management for Battery Swapping Station with an Intertemporal Framework

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Chen, Xinjiang Yang, Yu Wang, Jianxiao Song, Jie He, Guannan College of Engineering Peking University Beijing100091 China Department of Industrial and Systems Engineering University of Florida GainesvilleFL116595 United States National Engineering Laboratory for Big Data Analysis and Applications Peking University Beijing100091 China

Battery swapping as a business model for battery energy storage (BES) has great potential in future integrated low-carbon energy and transportation systems. However, frequent battery swapping will inevitably accelerate battery degradation and shorten the battery life accordingly. To model the tradeoff of BES use between energy and transportation applications coupled by battery swapping, we develop a life-cycle decision model that coordinates battery charging and swapping. This model is derived based on an improved intertemporal decision framework, in which the optimal marginal degradation cost (MDC) of BES is determined to maximize the BES benefit across time and application. The proposed framework and model are applied to manage a battery swapping station that simultaneously provides battery swapping services to electric vehicle customers and provides flexibility service to the power grid, including energy arbitrage and reserve. The case study shows that while the end of the physical life of BES occurs faster with battery swapping, the economic life becomes considerably longer. The results also reveal that the optimal MDC depends on the battery values in each application, and we analyze how the battery swapping price affects the optimal MDC and battery life. The proposed framework and model can also provide decision support for on-demand BES service, such as battery trading, renting and secondary use. Copyright © 2023, The Authors. All rights reserved.

关键词： Decision support systems

C olumnSGD: A Column-oriented Framework for Distributed Stochastic Gradient Descent

学校读者我要写书评

暂无评论

C olumnSGD: A Column-oriented Framework for Distributed Stoc...

International Conference on data engineering

作者： Zhipeng Zhang Wentao Wu Jiawei Jiang Lele Yu Bin Cui Ce Zhang School of EECS & Key Laboratory of High Confidence Software Technologies (MOE) Peking University Microsoft Research Redmond USA ETH Zurich Switzerland Tencent Inc Center for Data Science Peking University & National Engineering Laboratory for Big Data Analysis and Applications

ISBN: (数字)9781728129037

ISBN: (纸本)9781728129044

Distributed machine learning (ML) has triggered tremendous research interest in recent years. Stochastic gradient descent (SGD) is one of the most popular algorithms for training ML models, and has been implemented in almost all distributed ML systems, such as Spark MLlib, Petuum, MXNet, and TensorFlow. However, current implementations often incur huge communication and memory overheads when it comes to large models. One important reason for this inefficiency is the row-oriented scheme (RowSGD) that existing systems use to partition the training data, which forces them to adopt a centralized model management strategy that leads to vast amount of data exchange over the network. We propose a novel, column-oriented scheme (ColumnSGD) that partitions training data by columns rather than by rows. As a result, ML model can be partitioned by columns as well, leading to a distributed configuration where individual data and model partitions can be collocated on the same machine. Following this locality property, we develop a simple yet powerful computation framework that significantly reduces communication overheads and memory footprints compared to RowSGD, for large-scale ML models such as generalized linear models (GLMs) and factorization machines (FMs). We implement ColumnSGD on top of Apache Spark, and study its performance both analytically and experimentally. Experimental results on both public and real-world datasets show that ColumnSGD is up to 930× faster than MLlib, 63× faster than Petuum, and 14× faster than MXNet.

关键词： Conferences data engineering Radio frequency

PropNet: Propagating 2D Annotation to 3D Segmentation for Gastric Tumors on CT Scans

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Chen, Zifan Li, Jiazheng Zhao, Jie Liu, Yiting Li, Hongfeng Dong, Bin Tang, Lei Zhang, Li Center for Data Science Peking University Beijing China Radiology Department Peking University Cancer Hospital & Institute Beijing China National Engineering Laboratory for Big Data Analysis and Applications Peking University Beijing China Center for Health and Medical Data Science Peking University Beijing China Peking University Beijing China

Background: Accurate segmentation of gastric tumors on 3D CT scans is essential for diagnosis and treatment planning. However, irregular shapes and blurred boundaries of tumors make automatic segmentation challenging. Existing 2D-based methods do not model 3D morphology effectively, and 3D-based methods are computationally expensive and require annotated samples. Some studies have used few-shot learning (FSL) to train models with prior knowledge, but these methods cannot handle irregular shapes and blurred boundaries. Purpose: In this pilot study, we anticipate that the human-guided prior knowledge and specially designed modules will address the challenges of 3D segmentation of gastric tumors described above. Drawing inspiration from FSL, we aim to enable the model to learn from information propagation between slices rather than from specific segmentation annotation. Methods: We propose the PropNet framework to achieve 3D segmentation of gastric tumors by propagating the radiologist's prior knowledge from the 2D annotated slice into the entire 3D CT space. The proposing stage uses an inter-slice context module to extract contextual information between adjacent slices and generate a coarse segmentation. The refining stage then corrects difficult-to-segment pixels for improved segmentation. We use two-way branches for boundary and region in each stage to enhance the segmentation performance at tumor boundaries. Furthermore, we adopt an up-down parallel propagation strategy to speed up our model's predictions. Results: Our study utilizes 98 patients (173 CT scans) as a training set and 30 patients (59 CT scans) as a validation set. The proposed method achieves a comparable agreement with manual annotation (Dice of 0.803 with a p-value of 0.563) and radiologists (Dice of 0.797) on the validation set while significantly improving efficiency with only 40 seconds for annotation and segmentation of a tumor. Furthermore, we test our proposed method on an independent external

关键词： Refining

Nesterov's acceleration for approximate Newton

学校读者我要写书评

暂无评论

The Journal of Machine Learning Research

The Journal of Machine Learning Research 2020年第1期21卷 5627-5663页

作者： Haishan Ye Luo Luo Zhihua Zhang Shenzhen Research Insititution of Big Data The Chinese University of Hong Kong Shenzen Shenzhen China Department of Mathematics Hong Kong University of Science and Technology Kowloon Hong Kong National Engineering Lab for Big Data Analysis and Applications School of Mathematical Sciences Peking University Beijing China

Optimization plays a key role in machine learning. Recently, stochastic second-order methods have attracted considerable attention because of their low computational cost in each iteration. However, these methods might suffer from poor performance when the Hessian is hard to be approximate well in a computation-efficient way. To overcome this dilemma, we resort to Nesterov's acceleration to improve the convergence performance of these second-order methods and propose accelerated approximate Newton. We give the theoretical convergence analysis of accelerated approximate Newton and show that Nesterov's acceleration can improve the convergence rate. Accordingly, we propose an accelerated regularized sub-sampled Newton (ARSSN) which performs much better than the conventional regularized sub-sampled Newton empirically and theoretically. Moreover, we show that ARSSN has better performance than classical first-order methods empirically.

关键词： Nesterov's acceleration approximate Newton stochastic second-order

The Incoming Influenza Season—China,the United Kingdom,and the United States,2021–2022

学校读者我要写书评

暂无评论

China CDC weekly 2021年第49期3卷 1039-1045页

作者： Shasha Han Ting Zhang Yan Lyu Shengjie Lai Peixi Dai Jiandong Zheng Weizhong Yang Xiaohua Zhou Luzhao Feng Beijing International Center for Mathematical Research Peking UniversityBeijingChina Harvard Medical School Harvard UniversityBostonMAUSA School of Population Medicine and Public Health Chinese Academy of Medical Sciences&Peking Union Medical CollegeBeijingChina Academy for Advanced Interdisciplinary Studies Peking UniversityBeijingChina WorldPop School of Geography and Environmental ScienceUniversity of SouthamptonSouthamptonUK Division for Infectious Diseases Chinese Center for Disease Control and PreventionBeijingChina Department of Biostatistics School of Public HealthPeking UniversityBeijingChina National Engineering Laboratory of Big Data Analysis and Applied Technology Peking UniversityBeijingChina.

Introduction:Seasonal influenza activity has declined globally since the widespread of severe acute respiratory syndrome coronavirus 2(SARS-CoV-2)*** has been scarce information to understand the future dynamics of influenza—and under different hypothesis on relaxation of nonpharmaceutical interventions(NPIs)in particular—after the disruptions to seasonal ***:We collected data from public sources in China,the United Kingdom,and the United States,and forecasted the influenza dynamics in the incoming 2021–2022 season under different *** considered Northern China and Southern China separately,due to the sharp difference in the patterns of seasonal *** the United Kingdom,data were collected for England ***:Compared to the epidemics in 2017–2019,longer and blunter influenza outbreaks could occur should NPIs be fully lifted,with percent positivity varying from 10.5 to 18.6 in the studying *** rebounds would be smaller if the maskwearing intervention continued or the international mobility stayed low,but sharper if the mask-wearing intervention was lifted in the middle of influenza ***,influenza activity could stay low under a much less stringent mask-wearing intervention coordinated with influenza ***:The results added to our understandings of future influenza dynamics after the global decline during the coronavirus disease 2019(COVID-19)*** light of the uncertainty on the incoming circulation strains and the relatively low negative impacts of mask wearing on society,our findings suggested that wearing mask could be considered as an accompanying mitigation measure in influenza prevention and control,especially for seasons after long periods of low-exposure to influenza *** influenza activity declines globally during the coronavirus disease 2019(COVID-19)pandemic(1–4).For instance,in China,influenza activity,as measured by percentage of submitted specimens testing

关键词： prevention globally continued

Enabling and Scaling the HPCG Benchmark on the Newest Generation Sunway Supercomputer with 42 Million Heterogeneous Cores

学校读者我要写书评

暂无评论

Enabling and Scaling the HPCG Benchmark on the Newest Genera...

Supercomputing Conference

作者： Qianchao Zhu Hao Luo Chao Yang Mingshuo Ding Wanwang Yin Xinhui Yuan Center for Data Science Peking University Beijing China School of Mathematical Sciences Peking University Beijing China National Engineering Laboratory for Big Data Analysis and Applications Peking University Beijing China School of Electronic Engineering and Computer Science Peking University Beijing China National Research Center of Parallel Computer Engineering & Technology Beijing China

ISBN: (数字)9781450384421

ISBN: (纸本)9781665483902

We study and evaluate performance optimization techniques for the HPCG benchmark on the newest generation Sunway super-computer. Specifically, a two-level blocking scheme is proposed to expose adequate parallelism in the symmetric Gauss-Seidel kernel while keeping a fast convergence rate, a fine-grained kernel fusion technique is developed to alleviate the bandwidth load on local storage with small capacity, and a low overhead thread collaboration method is presented to efficiently move data between threads and hide its cost with data transfer operations. Test results show that the optimized HPCG code is able to exploit 73.0% of the theoretical memory bandwidth, and scale to over 42 million heterogeneous cores with 95.5% weak-scaling efficiency and 5.91 Pflop/s performance. We also study how the performance can be improved if the specific rules of HPCG are not fully obeyed, and design dependency preserving parallelization and vectorization methods, further boosting performance to 27.6 Pflop/ s.

关键词： Instruction sets High performance computing Memory management Bandwidth Benchmark testing Parallel processing data transfer

Enhance curvature information by structured stochastic quasi-Newton methods

学校读者我要写书评

暂无评论

arXiv 2020年

作者： Yang, Minghan Xu, Dong Chen, Hongyu Wen, Zaiwen Chen, Mengyun School of Mathematical Sciences Peking University China Beijing International Center for Mathematical Research Peking University China Center for Data Science Peking University China National Engineering Laboratory for Big Data Analysis and Applications Peking University China Huawei Technologies Co. Ltd China

In this paper, we consider stochastic second-order methods for minimizing a finite summation of nonconvex functions. One important key is to find an ingenious but cheap scheme to incorporate local curvature information. Since the true Hessian matrix is often a combination of a cheap part and an expensive part, we propose a structured stochastic quasi-Newton method by using partial Hessian information as much as possible. By further exploiting either the low-rank structure or the kronecker-product properties of the quasi-Newton approximations, the computation of the quasi-Newton direction is affordable. Global convergence to stationary point and local superlinear convergence rate are established under some mild assumptions. Numerical results on logistic regression, deep autoencoder networks and deep convolutional neural networks show that our proposed method is quite competitive to the state-of-the-art methods. Copyright © 2020, The Authors. All rights reserved.

关键词： Stochastic systems