Edge or Fog computing is emerging as a new computing paradigm where the data processing, networking, storage and analytics are performed closer to the devices (IoT) and applications. The edge of a network plays an important role in the IoT system. It is an optimal site for off-loading bandwidth hungry IoT data. In order to generate business value out of the large volume of data on the edge, we need decentralized machine learning (ML) algorithms. These algorithms must enable edge devices to communicate autonomously and deliver information seamlessly to the decision makers. In this paper, we present EdgeSGD a decentralized stochastic gradient descent method to solve a large linear regression problem on the edge network. The solution is applied to the seismic imaging use-case and evaluated using an edge computing testbed. The proposed algorithm avoids sending raw data to the cloud, and offers faster and balanced computation. We compare the algorithm with the existing methods such as MapReduce, DGD and EXTRA. The results show that the EdgeSGD converges faster to the optimal value and is robust towards node/link failure. Finally, we test the algorithm using real-world seismic data trace from Parkfield, CA and show that our proposed algorithm can illuminate the underlying fault region of San-Andreas.
1 March 2016 in Journal of Parallel and Distributed Computing
Existing volcano instrumentation and monitoring system use centralized approach for data collection and image reconstruction and they lack the capability of obtaining real time information. A new distributed method is required which can obtain a high resolution seismic tomography in real time. In this paper, we present a component-average distributed multiresolution evolving tomography algorithm for processing data and inverting volcano tomography in the network, while avoiding centralized computation and costly data collection. The new algorithm distributes the computational burden to sensor nodes and performs real time tomographic inversion under constraints of network resources. We implemented and evaluated the algorithm in a customized simulator using synthetic data. The experiment results validate that our proposed algorithm not only balances the computation load but also achieves high data loss tolerance.
Many real-world wireless sensor network applications such as environmental monitoring, structural health monitoring, and smart grid can be formulated as a least-squares problem. In distributed Cyber-Physical System (CPS), each sensor node observes partial phenomena due to spatial and temporal restriction and is able to form only partial rows of least-squares. Traditionally, these partial measurements were gathered at a centralized location. However, with the increase in sensors and their measurements, aggregation is becoming challenging and infeasible. In this paper, we propose distributed randomized kaczmarz that performs in-network computation to solve least-squares over the network by avoiding costly communication. As a case study, we present a volcano monitoring application on a distributed CORE emulator and use real data from Mt. St. Helens to evaluate our proposed method.
Tomography imaging, applied to seismology, requires a new, decentralized approach if high resolution calculations are to be performed in a sensor network configuration. The real-time data retrieval from a network of large-amount wireless seismic nodes to a central server is virtually impossible due to the sheer data amount and resource limitations. In this paper, we present a distributed multi-resolution evolving tomography algorithm for processing data and inverting volcano tomography in the network, while avoiding costly data collections and centralized computations. The new algorithm distributes the computational burden to sensor nodes and performs real-time tomography inversion under the constraints of network resources. We implemented and evaluated the system design in the CORE emulator. The experiment results validate that our proposed algorithm not only balances the computation load, but also achieves low communication cost and high data loss-tolerance.