Recent advances in System-on-Chip architectures have made the use of deep learning suitable for a number of applications on mobile devices. Unfortunately, due to the computational cost of neural network training, it i...
详细信息
ISBN:
(纸本)9781538649756
Recent advances in System-on-Chip architectures have made the use of deep learning suitable for a number of applications on mobile devices. Unfortunately, due to the computational cost of neural network training, it is often limited to inference task, e.g., prediction, on mobile devices. In this paper, we propose a deep learning framework that enables both deep learning training and inference tasks on mobile devices. While being able to accommodate withthe heterogeneity of computing devices technology on mobile devices, it also uses OpenCL to efficiently leverage modern SoC capabilities, e.g., multi-core CPU, integrated GPU and shared memory architecture, and accelerate deep learning computation. In addition, our system encodes the arithmetic operations of deep networks down to 8-bit fixed-point on mobile devices. As a proof of concept, we trained three well-known neural networks on mobile devices and exhibited a significant performance gain, energy consumption reduction, and memory saving.
the report covers details on such aspects of the JINR cloud development as migration to high availability setup based on Raft consensus algorithm, Ceph-based storage back-end for VM images and Dirac-based grid platfor...
详细信息
the proceedings contain 17 papers. the special focus in this conference is on Simulation and Modeling Methodologies,technologies and applications. the topics include: Agent-Based Modelling and Simulation Framework for...
ISBN:
(纸本)9783319698311
the proceedings contain 17 papers. the special focus in this conference is on Simulation and Modeling Methodologies,technologies and applications. the topics include: Agent-Based Modelling and Simulation Framework for Health Care;parameter Identification of Canalyzing and Nested Canalyzing Boolean Functions with Ternary Vectors for Gene Networks;the Power of Surrogate-Assisted Evolutionary computing in Searching Vaccination Strategy;modelling Population Dynamics Using a Hybrid Simulation Approach: Application to Healthcare;a Simulation-Based Dynamic Scheduling Method in Project Cost Estimation Process;future Prediction of Regional City Using Causal Inference Based on Time Series Data;Cooperative Radio Resources Allocation and Congestion Prevention Scheme for LTE-A;model-Based Development of a Multi-algorithm Harvest Planning System;cluster Performance Simulation for Spark Deployment Planning, Evaluation and Optimization;generator Platform of Benchmark Time-Lapsed Images Development of Cell Tracking Algorithms: Implementation of New Features Towards a Realistic Simulation of the Cell Spatial and Temporal Organization;proteins Flexibility as a Criterion for Elucidation of Activating Mutants in Personalized Cancer Medicine;distributed PowerShell Load Generator (D-PLG): A Tool for Generating Dynamic Network Traffic;HLogo: A Haskell STM-Based parallel Variant of NetLogo;requirements Gathering and Validation for Risk-Oriented Tool Support in Supply Chains;making Network Solvers Globally Convergent.
As distributed Energy Resources (DERs) are progressively integrated into electrical power systems, studies related to microgrid systems arise. Suitable testing tools are essential to accurately design and investigate ...
详细信息
ISBN:
(数字)9781728117225
ISBN:
(纸本)9781728117232
As distributed Energy Resources (DERs) are progressively integrated into electrical power systems, studies related to microgrid systems arise. Suitable testing tools are essential to accurately design and investigate microgrids and also to evaluate the corresponding control systems. this paper presents a testing platform for real-time simulation of microgrids with hardware-in-the-loop (HIL). A microgrid system with multiple DERs and loads is simulated in RTDS® real-time digital simulator. Typical control schemes for islanding detection, load shedding, and capacitor bank switching associated withthe islanded and grid-connected operation are implemented in the physical controller, which is interfaced with RTDS using several communication protocols. the presented testing platform proves the feasibility of real-time simulation for microgrid applications with flexibility for various testing scenarios.
Network traffic generation is one of the key technologies used to construct a network emulation environment for research into network protocols, applications and security mechanisms. In this paper, a new network traff...
详细信息
ISBN:
(纸本)9783030026981;9783030026974
Network traffic generation is one of the key technologies used to construct a network emulation environment for research into network protocols, applications and security mechanisms. In this paper, a new network traffic generator called NTGEDA is proposed that can generate network layer, application layer and user layer traffic on demand based on distributed agent architecture and the multi-level traffic model. the generator is evaluated by means of implementation. Experimental results illustrate that it is efficient in the network emulation environment for network protocols and security research.
this paper proposes that viable high-voltage high-power converters using wide band-gap SiC devices, will require the use of soft-switching, in particular zero-voltage switching techniques, if system parasitics and hig...
详细信息
ISBN:
(纸本)9784886864055
this paper proposes that viable high-voltage high-power converters using wide band-gap SiC devices, will require the use of soft-switching, in particular zero-voltage switching techniques, if system parasitics and high switching speeds are to be managed. An easy to understand and use quantitative comparison of different approaches can help us with selection of specific technologies for given applications. this paper presents the limitations of current approaches, and proposes a new converter figure of merit (CFOM) as the basis for understanding the limitations of high power converters based on wide band-gap SiC devices, showing that soft-switching is the key.
Software packages, created for the modern physics experiments, present a sets intertwined code structures, written by different people and bundled together to perform range of different reconstruction and analysis tas...
详细信息
the paper presents a new development in the Multipoint Approximation Method (MAM) that makes it capable of handling large-scale problems. the approach relies on approximations built in the space of design variables wi...
详细信息
ISBN:
(纸本)9783319996738;9783319996721
the paper presents a new development in the Multipoint Approximation Method (MAM) that makes it capable of handling large-scale problems. the approach relies on approximations built in the space of design variables within the iterative trust-region-based framework of MAM. Withthe purpose of solving high dimensionality problems in a reasonable time, a parallel variant of the Multipoint Approximation Method (PMAM) has been developed. It is supposed that the values of the objective function and those of the constraints are computed using distributed memory (on several cluster nodes), whereas the optimization module runs on a single node using shared memory. Numerical experiments have been carried out on a benchmark example of structural optimization.
this paper introduces HSAStore, a newly designed storage system whose goal goes to build an efficient storage system for computing systems that containing large-scale intermediate data. HSAStore involves three sub-sys...
详细信息
ISBN:
(纸本)9783030009168;9783030009151
this paper introduces HSAStore, a newly designed storage system whose goal goes to build an efficient storage system for computing systems that containing large-scale intermediate data. HSAStore involves three sub-systems to work together, a distributed file system which is in charge of the intermediate data, a centralized Network Attached Storage (NAS) which stores the raw input data and the results data, and a local file system that serves for the local data. the HSAStore takes full advantage of the network bandwidth of the computing cluster. Moreover, HSAStore adopts the distributed file system, so it is helpful for efficient execution of parallel programs. Experiments show that HSAStore has a significant improvement on efficiency in the computing systems that containing large-scale intermediate data.
the main directions of development of the data center of the Institute of Physics of NAS of Azerbaijan are grid and cloud technologies. the grid site of the data center is integrated into the EGI/WLCG and ATLAS (CERN)...
详细信息
暂无评论