Engineering Commons^™

Amodal Instance Segmentation And Multi-Object Tracking With Deep Pixel Embedding, Yanfeng Liu

Department of Electrical and Computer Engineering: Dissertations, Theses, and Student Research

This thesis extends upon the representational output of semantic instance segmentation by explicitly including both visible and occluded parts. A fully convolutional network is trained to produce consistent pixel-level embedding across two layers such that, when clustered, the results convey the full spatial extent and depth ordering of each instance. Results demonstrate that the network can accurately estimate complete masks in the presence of occlusion and outperform leading top-down bounding-box approaches.

The model is further extended to produce consistent pixel-level embeddings across two consecutive image frames from a video to simultaneously perform amodal instance segmentation and multi-object tracking. No post-processing …

Domain Adaptation In Unmanned Aerial Vehicles Landing Using Reinforcement Learning, Pedro Lucas Franca Albuquerque

Department of Computer Science and Engineering: Dissertations, Theses, and Student Research

Landing an unmanned aerial vehicle (UAV) on a moving platform is a challenging task that often requires exact models of the UAV dynamics, platform characteristics, and environmental conditions. In this thesis, we present and investigate three different machine learning approaches with varying levels of domain knowledge: dynamics randomization, universal policy with system identification, and reinforcement learning with no parameter variation. We first train the policies in simulation, then perform experiments both in simulation, making variations of the system dynamics with wind and friction coefficient, then perform experiments in a real robot system with wind variation. We initially expected that providing …

Data Analytics And Machine Learning To Enhance The Operational Visibility And Situation Awareness Of Smart Grid High Penetration Photovoltaic Systems, Aditya Sundararajan

FIU Electronic Theses and Dissertations

Electric utilities have limited operational visibility and situation awareness over grid-tied distributed photovoltaic systems (PV). This will pose a risk to grid stability when the PV penetration into a given feeder exceeds 60% of its peak or minimum daytime load. Third-party service providers offer only real-time monitoring but not accurate insights into system performance and prediction of productions. PV systems also increase the attack surface of distribution networks since they are not under the direct supervision and control of the utility security analysts.

Six key objectives were successfully achieved to enhance PV operational visibility and situation awareness: (1) conceptual cybersecurity …

Stochastic Resonance Enables Bpp/Log∗ Complexity And Universal Approximation In Analog Recurrent Neural Networks, Emmett Redd, A. Steven Younger, Tayo Obafemi-Ajayi

Electrical and Computer Engineering Faculty Research & Creative Works

Stochastic resonance (SR) is a natural process that without limit increases the precision of signal measurements in biological and physical sciences. Most artificial neural networks (NNs) are implemented on digital computers of fixed precision. A NN accessing universal approximation and a computational complexity class more powerful that of a Turing machine needs analog signals utilizing SR's limitless precision increase. This paper links an analog recurrent (AR) NN theorem, SR, BPP/log∗ (a physically realizable, super-Turing computation class), and universal approximation so NNs following them can be made computationally more powerful. An optical neural network mimicking chaos indicates super-Turing computation has been …

Simulation Of Human Balance Control Using An Inverted Pendulum Model, Joshua E. Caneer

Undergraduate Research & Mentoring Program

The nervous system that human beings use to control balance is remarkably adaptable to a wide variety of environments and conditions. This neural system is likely a combination of many inputs and feedback control loops working together. The ability to emulate this system of balance could be of great value in understanding and developing solutions to proprioceptive disorders and other diseases that affect the human balance control system. Additionally, the process of emulating the human balance system may also have widespread applications to the locomotion capabilities of many types of robots, in both bipedal and non-bipedal configurations.

The goal of …

Depressiongnn: Depression Prediction Using Graph Neural Network On Smartphone And Wearable Sensors, Param Bidja

Honors Scholar Theses

Depression prediction is a complicated classification problem because depression diagnosis involves many different social, physical, and mental signals. Traditional classification algorithms can only reach an accuracy of no more than 70% given the complexities of depression. However, a novel approach using Graph Neural Networks (GNN) can be used to reach over 80% accuracy, if a graph can represent the depression data set to capture differentiating features. Building such a graph requires 1) the definition of node features, which must be highly correlated with depression, and 2) the definition for edge metrics, which must also be highly correlated with depression. In …

Application Of Artificial Neural Networks To Assess Student Happiness, Gokhan Egilmez, Nadiye O. Erdil, Omid Mohammadi Arani, Mana Vahid

Mechanical and Industrial Engineering Faculty Publications

The purpose of this study is to develop an analytical assessment approach to identify the main factors that affect graduate students' happiness level. The two methods, multiple linear regression (MLR) and artificial neural networks (ANN), were employed for analytical modelling. A sample of 118 students at a small non-profit private university constituted the survey pool. Various factors including education, school facilities, health, social activities, and family were taken into consideration as a result of literature review in happiness assessment. A total of 32 inputs and one output variables were identified during survey design phase. The following survey conduction, data collection, …

Feature Extraction Using Spiking Convolutional Neural Networks, Ruthvik Vaila, John Chiasson, Vishal Saxena

Electrical and Computer Engineering Faculty Publications and Presentations

Spiking neural networks are biologically plausible counterparts of the artificial neural networks, artificial neural networks are usually trained with stochastic gradient descent and spiking neural networks are trained with spike timing dependant plasticity. Training deep convolutional neural networks is a memory and power intensive job. Spiking networks could potentially help in reducing the power usage. There is a large pool of tools for one to chose to train artificial neural networks of any size, on the other hand all the available tools to simulate spiking neural networks are geared towards computational neuroscience applications and they are not suitable for real …

A Statistical Approach To Provide Explainable Convolutional Neural Network Parameter Optimization, Saman Akbarzadeh, Selam Ahderom, Kamal Alameh

Research outputs 2014 to 2021

Algorithms based on convolutional neural networks (CNNs) have been great attention in image processing due to their ability to find patterns and recognize objects in a wide range of scientific and industrial applications. Finding the best network and optimizing its hyperparameters for a specific application are central challenges for CNNs. Most state-of-the-art CNNs are manually designed, while techniques for automatically finding the best architecture and hyperparameters are computationally intensive, and hence, there is a need to severely limit their search space. This paper proposes a fast statistical method for CNN parameter optimization, which can be applied in many CNN applications …

An Explainable Autoencoder For Collaborative Filtering Recommendation, Pegah Sagheb Haghighi, Olurotimi Seton, Olfa Nasraoui

Faculty Scholarship

Autoencoders are a common building block of Deep Learning architectures, where they are mainly used for representation learning. They have also been successfully used in Collaborative Filtering (CF) recommender systems to predict missing ratings. Unfortunately, like all black box machine learning models, they are unable to explain their outputs. Hence, while predictions from an Autoencoderbased recommender system might be accurate, it might not be clear to the user why a recommendation was generated. In this work, we design an explainable recommendation system using an Autoencoder model whose predictions can be explained using the neighborhood based explanation style. Our preliminary work …