distributed stochastic gradient descent (SGD) is essential for scaling the machine learning algorithms to a large number of computing nodes. However, the infrastructures variability such as high communication delay or...
详细信息
ISBN:
(纸本)9781509066315
distributed stochastic gradient descent (SGD) is essential for scaling the machine learning algorithms to a large number of computing nodes. However, the infrastructures variability such as high communication delay or random node slowdown greatly impedes the performance of distributed SGD algorithm, especially in a wireless system or sensor networks. In this paper, we propose an algorithmic approach named Overlap Local-SGD (and its momentum variant) to overlap communication and computation so as to speedup the distributed training procedure. The approach can help to mitigate the straggler effects as well. We achieve this by adding an anchor model on each node. After multiple local updates, locally trained models will be pulled back towards the synchronized anchor model rather than communicating with others. Experimental results of training a deep neural network on CIFAR-10 dataset demonstrate the effectiveness of Overlap Local-SGD. We also provide a convergence guarantee for the proposed algorithm under non-convex objective functions.
The communication between vehicles has attracted more and more attention in intelligent transportation system(ITS). Due to the rapid movement of vehicles, the topology of network nodes changes rapidly, resulting in co...
详细信息
The use of data is increasing steadily in the modern era of technology. That produced the term big data. Many companies are interested in analyzing this data, which amounts to several terabytes. The term distributed c...
详细信息
Asynchronous Stochastic computing (ASC) is a branch of clockless Stochastic computing methodology that encodes signals as a digital asynchronous pulse-width modulated stream that carries information within it's du...
详细信息
Asynchronous Stochastic computing (ASC) is a branch of clockless Stochastic computing methodology that encodes signals as a digital asynchronous pulse-width modulated stream that carries information within it's duty cycle and frequency. In this paper a comprehensive study into the energy and frequency optimization of Asynchronous Stochastic computing circuits is presented. Design knobs that affect characteristics of the stream are identified and, by evaluating their sensitivity, a Pareto optimization strategy is derived. In distributedcomputingsystems such as wearable sensors, dynamic throughput scaling is often required. The insights from the Pareto analysis, are utilized to design a scalable throughput Asynchronous Stochastic computing Arithmetic unit, that is capable of performing Multiply-Accumulate (MAC) operations. Our simulated results in the design of the arithmetic unit prove the effectiveness of the single optimization problem through an average energy savings of 17- 32% across the two different throughput regions of operation.
We present an approach for enabling a distributed anonymization process over large collections of sensor data. Our approach anonymizes large datasets (which might not fit in main memory) using an arbitrary number of w...
详细信息
ISBN:
(纸本)9781665404242;9781665447249
We present an approach for enabling a distributed anonymization process over large collections of sensor data. Our approach anonymizes large datasets (which might not fit in main memory) using an arbitrary number of workers within the Spark framework. We describe how to parallelize the anonymization process through a proper partitioning of the dataset. Our experimental evaluation shows that the proposed approach is scalable and do not affect the quality of the anonymized dataset.
A severe existential danger to agriculture is being posed by extreme weather patterns and falling forests, which require an evolution of current agricultural systems. Traditional agriculture can become smart agricultu...
详细信息
ISBN:
(纸本)9781665491129
A severe existential danger to agriculture is being posed by extreme weather patterns and falling forests, which require an evolution of current agricultural systems. Traditional agriculture can become smart agriculture with the aid of the Internet of Things. sensor-based monitoring systems for agriculture can benefit farmers in the long run. A network of physical things called the “Internet of Things” uses networking and sensors to enhance data collecting and exchange. IoT networks, which are less expensive and use less energy, can be distributed over a large population region. We propose an energy-efficient, low-cost wireless network for smart agriculture in this article. Our paper aims to develop a new hybrid topology that is both stable and scalable; additionally, nodes will self-conFigure to form the topology without the need for manual interference. In order to connect to the Internet, we used NodeMCU, a system-on-chip (SOC) that combines a Tensilica L106 32-bit micro controller unit with a Wi-Fi transceiver. By implementing UPnP for our NodeMCUs linked with a tree topology Wi-Fi mesh, we were able to build a multicast tree algorithm. These NodeMCUs will self-discover via SSDP (Simple Service Discovery Protocol), sending a NOTIFY request to the multicast IP on the link, and the master node will determine the position. The mechanism underlying automatic topology forming and node configuration is explored in depth in this paper. This algorithm and topology enables energy efficient and easy to implement sensor networks, which can help in better collection of data, in turn help farmers take informed decisions.
This paper demonstrates a portable pH sensing system which converts pH to a digital readout. The system consists of a CMOS ISFET pH sensor with discrete analog to digital conversion and display modules. The front end ...
详细信息
ISBN:
(纸本)9781728180588
This paper demonstrates a portable pH sensing system which converts pH to a digital readout. The system consists of a CMOS ISFET pH sensor with discrete analog to digital conversion and display modules. The front end sensor was fabricated in standard 0.5 mu m CMOS technology and an easy to use Arduino microcontroller is used as the computing device. Measurement results have reasonable agreements with the simulations. The average sensitivity in simulation is 46 mV/pH and it is 42 mV/pH based on experimentally extracted capacitance parameters. The design uses a differential ISFET readout circuit with 12 mu W of power consumption at 2 V and 80 mu W of power consumption at 5 V. The portable sensor implemented is 10cm x 16cm, while the main ISFET front end readout occupies a chip area of 0.0093 mm(2). This portable device is suitable for environmental and biomedical applications.
Decentralized systems, such as distributed ledgers and the InterPlanetary File System (IPFS), are designed to offer more open and robust services. However, they also create opportunities for illegal activities. We dem...
详细信息
ISBN:
(纸本)9781728166803
Decentralized systems, such as distributed ledgers and the InterPlanetary File System (IPFS), are designed to offer more open and robust services. However, they also create opportunities for illegal activities. We demonstrate how these technologies can be used to launch a ransomware as a service campaign. We show that criminals can transact with affiliates and victims without having to reveal their identity. Furthermore, by exploiting the robustness and resilience to churn of IPFS, as well as the decentralized computing capabilities of Ethereum, criminals can remain offline during most procedures, with many privacy guarantees.
This paper focuses on the weighted set cover problem in networking systems and presents a fully distributed algorithm from the perspective of Nash equilibrium learning and selection. By viewing each set as an agent, w...
详细信息
ISBN:
(数字)9781665467612
ISBN:
(纸本)9781665467629
This paper focuses on the weighted set cover problem in networking systems and presents a fully distributed algorithm from the perspective of Nash equilibrium learning and selection. By viewing each set as an agent, we recast the problem as a networked ordinal potential game and classify the resulting Nash equilibrium into two categories. We show that each inferior Nash equilibrium (INE) could always be improved via local action exchange and better approximations could be achieved via self-organized selection among superior Nash equilibria (SNEs). By showing the existence of an improvement path that leads any action profile to an SNE, we prove that our algorithm converges in finite time to a conventional Nash equilibrium, where the joint action is a selected SNE. Comparison experiments with typical methods demonstrate the superiority to the state of the art.
The article proposes the use of a formal approach to the decomposition of the architecture of a decentralized coordination control system of distributed cyber-physical systems using microservices, which minimizes the ...
详细信息
ISBN:
(纸本)9781665426060
The article proposes the use of a formal approach to the decomposition of the architecture of a decentralized coordination control system of distributed cyber-physical systems using microservices, which minimizes the cost of its creation and maintenance. This approach is based on the convertion of system gipergraph to optimal bipartite graph by the criterium of minimize interconnection flows and modification of Stoer-Wagner algorithm.
暂无评论