Node architectures of extreme-scale systems are rapidly increasing in complexity. Emerging homogenous and heterogeneous designs provide massive multi-level parallelism, but developing efficient runtime systems and mid...
详细信息
Node architectures of extreme-scale systems are rapidly increasing in complexity. Emerging homogenous and heterogeneous designs provide massive multi-level parallelism, but developing efficient runtime systems and middleware that allow applications to efficiently and productively exploit these architectures is extremely challenging. Moreover, current state-of-the-art approaches may become unworkable once energy consumption, resilience, and data movement constraints are added. The goal of this workshop is to attract the international research community to share new and bold ideas that will address the challenges of design, implementation, deployment, and evaluation of future runtime systems and middleware.
Federated Learning is a distributed model training system that uses the computing resources and private data of participants to collaboratively train Machine Learning models. In order to incentivize data holders to ac...
详细信息
ISBN:
(数字)9798331509712
ISBN:
(纸本)9798331509729
Federated Learning is a distributed model training system that uses the computing resources and private data of participants to collaboratively train Machine Learning models. In order to incentivize data holders to actively participate in Federated Learning, a crucial issue is how to fairly assess each participant’s contribution. One class of the mainstream methods estimates client contributions by calculating the correlation between local and global model parameters, substantially reducing the reliance on large computing resources and test datasets. However, these methods overlook an implicit key factor, leading to inaccuracy in assessing the contribution of the participants. We propose FedCA, which provides a fairness assessment of the client’s contribution based on the performance gain of the global model in each communication round. The experimental results demonstrate that FedCA accurately identifies the contributions of clients with varying data qualities and effectively approximates the optimal fairness method of Shapley values. The source code for this paper is available at https://***/paper-liu/***.
A new learning algorithm for multilayer neural networks is introduced. The algorithm merges the gradient technique and least squares technique. The presented algorithm is compared with backpropagation learning rule on...
详细信息
A new learning algorithm for multilayer neural networks is introduced. The algorithm merges the gradient technique and least squares technique. The presented algorithm is compared with backpropagation learning rule on the parity problem. Simulation results are presented.
The paper is devoted to analysis of a strategy of computation distribution on heterogeneous parallel systems. According to this strategy processes of parallel program are distributed over the processors according to t...
详细信息
The paper is devoted to analysis of a strategy of computation distribution on heterogeneous parallel systems. According to this strategy processes of parallel program are distributed over the processors according to their performances and data are distributed between processes evenly. The paper presents an algorithm that computes optimal number of the processes and their distribution over processors minimizing the execution time of an application. The processor performance is considered as a function of the number of processes running on the processor and the amount of the data processing by the processor.
Online and telecommunication services reduce the gap between level of health care service quality between villages and cities. In this research work, a study of five e-health models has been done. These models are sim...
详细信息
ISBN:
(纸本)9781728139890
Online and telecommunication services reduce the gap between level of health care service quality between villages and cities. In this research work, a study of five e-health models has been done. These models are simultaneously in use these days in developing countries such as India. The models are discussed in terms of their evolution and their application in various kinds of eHealth services. The paper gives definitive information on the technical standards and stacks required to build such eHealth communication models and services. Our study has found that most of these models are running concurrently in most of the developing countries and there is no absolute winner when they are compared with each other. It all depends upon the level of development in the respective country and commitment of their governments to provide affordable health care services to the masses. The paper also discusses the use of Cloud-based medical sensors, IoT and Fog computing approach as well the use of data mining and machine learning modelling to construct an e-Health System. In the second part of this paper, mathematical models and algorithms to diagnose the onset of health problems are given. The outcome of this study shows that there is a need for statistical as well as machine learning models to build eHealth solutions in developing countries. After this research a new model has been proposed which aims at providing better and affordable healthcare services and coincidentally benefits various other sectors.
DAOS is an open-source scale-out object store designed from the ground up to deliver extremely high bandwidth/IOPS and low latency I/Os to the most demanding data-intensive workloads. It aims at supporting nextgen sci...
详细信息
ISBN:
(数字)9781665497473
ISBN:
(纸本)9781665497480
DAOS is an open-source scale-out object store designed from the ground up to deliver extremely high bandwidth/IOPS and low latency I/Os to the most demanding data-intensive workloads. It aims at supporting nextgen scientific workflows combining simulation, big data and AI in a single storage tier. DAOS presents a rich and scalable storage interface that allows efficient storage of both structured and unstructured data. DAOS supports multiple application interfaces including a parallel filesystem, Hadoop/Spark connector, TensorFlow-IO, native Python dictionary bindings, HDF5, MPI-IO as well as domain-specific data models like SEGY. Many DAOS deployments are underway including a 230PB installation connected to the ALCF's Aurora system and a 1PB DAOS system for LRZ's SuperMUC-NG phase 2.
This paper studies a class of source coding problems that combines elements of the CEO problem with the multiple description problem. In this setting, noisy versions of one remote source are observed by two nodes with...
详细信息
This paper studies a class of source coding problems that combines elements of the CEO problem with the multiple description problem. In this setting, noisy versions of one remote source are observed by two nodes with encoders (which is similar to the CEO problem). However, it differs from the CEO problem in that each node must generate multiple descriptions of the source. This problem is of interest in multiple scenarios in efficient communication over networks. In this paper, an achievable region and an outer bound are presented for this problem, which is shown to be sum rate optimal for a class of distortion constraints.
It is suggested to use for construction of the computer systems by the reconfiguration architecture of topology on the base of memory of collective access that has a high system intellect and regular structure. It is ...
详细信息
It is suggested to use for construction of the computer systems by the reconfiguration architecture of topology on the base of memory of collective access that has a high system intellect and regular structure. It is offered wide use of network architectures with reconfiguration topology in case of the distributed processing of data.
The paper describes a platform support environment layer proposed by the TMN Computing Platform Special Interest Group (RACE programme) whose main objective is to provide enough support for the contrasting TMN managem...
详细信息
The paper describes a platform support environment layer proposed by the TMN Computing Platform Special Interest Group (RACE programme) whose main objective is to provide enough support for the contrasting TMN management applications. The paper also discusses the IDEA project proposed by the PRISM Laboratory. It has developed a platform support environment for network management systems. To finish, the authors show the IDEA platform support environment and its relationship with this general TMN computing platform architecture. The important services supported by both the implementation developed by the IDEA project and the CPSIG architecture are identified and discussed.< >
暂无评论