Wang, Dong

Profile Picture

Filters

2018 - 201912020 - 20231
Reset filters

Settings

Citations

Publication Search Results

Now showing 1 - 2 of 2
  • Thumbnail Image
    Publication
    CFL-HC: A coded federated learning framework for heterogeneous computing scenarios
    (2023-07-07) Wang, Dong; Lu, Kejie; College of Engineering; Rodríguez Martínez, Manuel; Arzuaga Cruz, Emmanuel; Rivera Gallego, Wilson; Department of Computer Science and Engineering; Rodriguez Román, Daniel
    Federated learning (FL) is a promising machine learning paradigm that enables distributed edge devices to collaboratively train a model without sharing their raw data. However, a major challenge in FL is that edge devices are heterogeneous, which compromises the convergence rate of model training. To mitigate this influence, several recent studies have proposed various solutions, one of which is to utilize coded computing to facilitate the training of linear models. Nevertheless, the existing coded federated learning (CFL) scheme is limited by a fixed coding redundancy parameter. Besides, a weight matrix used in the existing design may introduce unnecessary errors. To tackle these limitations, we propose a novel framework to facilitate CFL model training in heterogeneous computing scenarios. Our framework applies a computing system consisting of a central server and multiple computing devices with original or coded datasets. With specifying an expected number of input-output pairs used in one round, we formulate an optimization problem to determine the best deadline for each training round and the optimal size of the computing task allocated to each computing device. To optimize this problem, we design a two-step alternative solution and evaluate the proposed framework by developing a real CFL system using the platform of message-passing interface (MPI). By conducting numerical experiments, we demonstrate the advantages of our framework in terms of both accuracy and convergence speed. Besides, to train a model using real-world data, we apply a kernel method and encoding technique to transform the nonlinear data sample into data pairs with linear properties. We build a distributed system to test the performance of the proposed scheme. The analysis of the experiment shows consistent results.
  • Thumbnail Image
    Publication
    Towards the development of an automated, portable enterococcus faecalis biosensor for recreational waters
    (2018) Wang, Dong; Resto-Irizarry, Pedro J.; College of Engineering; Díaz-Rivera, Rubén E.; Rodríguez-Abudo, Sylvia; Department of Mechanical Engineering; Otero-Morales, Ernesto
    The quality of recreational waters is of great importance to the public. Enterococcus faecalis is one of the main indicators of recreational water quality. The standardized technology for quantifying bacterial concentrations require the transportation of water samples to a lab and measurement of the test by trained personnel. These actions delay the publishing of results from water quality tests. Thus, a portable, automated, and in-situ water quality solution can yield advantages towards the public welfare. In this project, we design and build a novel biosensor that is portable and sensitive to various bacterial concentrations. We use rapid prototyping techniques to build and automate an enzyme-based water quality assay using a millifluidic biosensor. We use a Raspberry Pi micro-computer, a CMOS image sensor, an Arduino micro-controller and Python programing for control and data acquisition. Fluorescence from the enzyme-based assay is indicative of bacterial growth and is measured over a 24-hour period. Our sensor has detected a bacterial concentration as low as 23 Colony Forming Units (CFU) per 100 milliliters. The CMOS camera and RGB analysis is used to measure the increase of fluorescence from the assay as a function of time. Blue light is indicative of fluorescence from the assay and of a positive contamination result.