Engineering Commons^™

Toward Generating Efficient Deep Neural Networks, Chengcheng Li

Doctoral Dissertations

Recent advances in deep neural networks have led to tremendous applications in various tasks, such as object classification and detection, image synthesis, natural language processing, game playing, and biological imaging. However, deploying these pre-trained networks on resource-limited devices poses a challenge, as most state-of- the-art networks contain millions of parameters, making them cumbersome and slow in real-world applications. To address this problem, numerous network compression and acceleration approaches, also known as efficient deep neural networks or efficient deep learning, have been investigated, in terms of hardware and software (algorithms), training, and inference. The aim of this dissertation is to study …

Development Of Flood Prediction Models Using Machine Learning Techniques, Bhanu Kanwar

Doctoral Dissertations

"Flooding and flash flooding events damage infrastructure elements and pose a significant threat to the safety of the people residing in susceptible regions. There are some methods that government authorities rely on to assist in predicting these events in advance to provide warning, but such methodologies have not kept pace with modern machine learning. To leverage these algorithms, new models must be developed to efficiently capture the relationships among the variables that influence these events in a given region. These models can be used by emergency management personnel to develop more robust flood management plans for susceptible areas. The research …

Theoretical And Experimental Application Of Neural Networks In Spaceflight Control Systems, Pavel Galchenko

Doctoral Dissertations

“Spaceflight systems can enable advanced mission concepts that can help expand our understanding of the universe. To achieve the objectives of these missions, spaceflight systems typically leverage guidance and control systems to maintain some desired path and/or orientation of their scientific instrumentation. A deep understanding of the natural dynamics of the environment in which these spaceflight systems operate is required to design control systems capable of achieving the desired scientific objectives. However, mitigating strategies are critically important when these dynamics are unknown or poorly understood and/or modelled. This research introduces two neural network methodologies to control the translation and rotation …

Development Of Horizontal Axis Hydrokinetic Turbine Using Experimental And Numerical Approaches, Abdulaziz Abutunis

Doctoral Dissertations

“Hydrokinetic energy conversion systems (HECSs) are emerging as viable solutions for harnessing the kinetic energy in river streams and tidal currents due to their low operating head and flexible mobility. This study is focused on the experimental and numerical aspects of developing an axial HECS for applications with low head ranges and limited operational space. In Part I, blade element momentum (BEM) and neural network (NN) models were developed and coupled to overcome the BEM’s inherent convergence issues which hinder the blade design process. The NNs were also used as a multivariate interpolation tool to estimate the blade hydrodynamic characteristics …

Neuroengineering Of Clustering Algorithms, Leonardo Enzo Brito Da Silva

Doctoral Dissertations

"Cluster analysis can be broadly divided into multivariate data visualization, clustering algorithms, and cluster validation. This dissertation contributes neural network-based techniques to perform all three unsupervised learning tasks. Particularly, the first paper provides a comprehensive review on adaptive resonance theory (ART) models for engineering applications and provides context for the four subsequent papers. These papers are devoted to enhancements of ART-based clustering algorithms from (a) a practical perspective by exploiting the visual assessment of cluster tendency (VAT) sorting algorithm as a preprocessor for ART offline training, thus mitigating ordering effects; and (b) an engineering perspective by designing a family of …

Neuron Clustering For Mitigating Catastrophic Forgetting In Supervised And Reinforcement Learning, Benjamin Frederick Goodrich

Doctoral Dissertations

Neural networks have had many great successes in recent years, particularly with the advent of deep learning and many novel training techniques. One issue that has affected neural networks and prevented them from performing well in more realistic online environments is that of catastrophic forgetting. Catastrophic forgetting affects supervised learning systems when input samples are temporally correlated or are non-stationary. However, most real-world problems are non-stationary in nature, resulting in prolonged periods of time separating inputs drawn from different regions of the input space.

Reinforcement learning represents a worst-case scenario when it comes to precipitating catastrophic forgetting in neural networks. …

Approximate Dynamic Programming Based Solutions For Fixed-Final-Time Optimal Control And Optimal Switching, Ali Heydari

Doctoral Dissertations

"Optimal solutions with neural networks (NN) based on an approximate dynamic programming (ADP) framework for new classes of engineering and non-engineering problems and associated difficulties and challenges are investigated in this dissertation. In the enclosed eight papers, the ADP framework is utilized for solving fixed-final-time problems (also called terminal control problems) and problems with switching nature. An ADP based algorithm is proposed in Paper 1 for solving fixed-final-time problems with soft terminal constraint, in which, a single neural network with a single set of weights is utilized. Paper 2 investigates fixed-final-time problems with hard terminal constraints. The optimality analysis of …