Machine Learning-Based Asynchronous Computational Framework for Generalized Kalman Filter

Date of Award

8-2021

Degree Name

Doctor of Philosophy

Department

Electrical and Computer Engineering

First Advisor

Dr. Ikhlas Abdel-Qader

Second Advisor

Dr. Janos Grantner

Third Advisor

Dr. Azim Houshyar

Keywords

Asynchronous Kalman Filter, machine learning based thread classifier, OpenMP Thread Classifier, GPU Thread Classifier, asynchronous computation, CUDA and OpenMP Message Passing

Abstract

Kalman filter plays a significant role in most modern dynamic systems from continuously monitoring minuscule sensors to tracking and positioning of largest aircrafts and ships. Although the Kalman filter algorithms are well suited to be executed on most digital systems, they become slow when applied to large-scale dynamic systems—systems with vectors of high dimensions. Therefore, efficient execution of Kalman filter for the time-critical and large-scale applications is of the essence. This study aims to address this necessity by improving the computational efficiency of Kalman filter in real-time applications. This work develops a novel framework to improve the performance of a generalized Kalman filter with unknown inputs (GKF-UI) using multithreaded-multicore processors and machine learning (ML) classification methods. An asynchronous execution model based on OpenMP shared-memory message-passing framework is developed and integrated with a novel supervised learning-based thread classifier for the GKF-UI algorithm to enhance its computational efficiency. The experimental results show that the proposed approach can achieve up to 35x speedup over the serial single-threaded implementations and can play a significant role in large-scale systems as well as for the time-critical applications. The ML-based asynchronous GKF-UI is also implemented in this work to predict blood glucose concentration as a time-critical application. Results demonstrated accurate prediction, based on the distance between the observed and predicted values, with speedup up to 2.8x over those recently reported methods. The proposed novel machine learning-based thread-classifier and asynchronous computational framework can be used in a wide array of massively parallel architectures such as GPUs to enhance performance of computationally intensive algorithms

Access Setting

Dissertation-Abstract Only

Restricted to Campus until

8-15-2031

This document is currently not available here.

Share

COinS