Computer Sciences Commons^™

Improving Space Efficiency Of Deep Neural Networks, Aliakbar Panahi

Theses and Dissertations

Language models employ a very large number of trainable parameters. Despite being highly overparameterized, these networks often achieve good out-of-sample test performance on the original task and easily fine-tune to related tasks. Recent observations involving, for example, intrinsic dimension of the objective landscape and the lottery ticket hypothesis, indicate that often training actively involves only a small fraction of the parameter space. Thus, a question remains how large a parameter space needs to be in the first place — the evidence from recent work on model compression, parameter sharing, factorized representations, and knowledge distillation increasingly shows that models can be …

Visualization For Solving Non-Image Problems And Saliency Mapping, Divya Chandrika Kalla

All Master's Theses

High-dimensional data play an important role in knowledge discovery and data science. Integration of visualization, visual analytics, machine learning (ML), and data mining (DM) are the key aspects of data science research for high-dimensional data. This thesis is to explore the efficiency of a new algorithm to convert non-images data into raster images by visualizing data using heatmap in the collocated paired coordinates (CPC). These images are called the CPC-R images and the algorithm that produces them is called the CPC-R algorithm. Powerful deep learning methods open an opportunity to solve non-image ML/DM problems by transforming non-image ML problems into …

On Studying Distributed Machine Learning, Simeon Eberz

Senior Honors Theses

The Internet of Things (IoT) is utilizing Deep Learning (DL) for applications such as voice or image recognition. Processing data for DL directly on IoT edge devices reduces latency and increases privacy. To overcome the resource constraints of IoT edge devices, the computation for DL inference is distributed between a cluster of several devices. This paper explores DL, IoT networks, and a novel framework for distributed processing of DL in IoT clusters. The aim is to facilitate and simplify deployment, testing, and study of a distributed DL system, even without physical devices. The contributions of this paper are a deployment …

Scaling Up Exact Neural Network Compression By Relu Stability, Thiago Serra, Xin Yu, Abhinav Kumar, Srikumar Ramalingam

Faculty Conference Papers and Presentations

We can compress a rectifier network while exactly preserving its underlying functionality with respect to a given input domain if some of its neurons are stable. However, current approaches to determine the stability of neurons with Rectified Linear Unit (ReLU) activations require solving or finding a good approximation to multiple discrete optimization problems. In this work, we introduce an algorithm based on solving a single optimization problem to identify all stable neurons. Our approach is on median 183 times faster than the state-of-art method on CIFAR-10, which allows us to explore exact compression on deeper (5 x 100) and wider …

Computational Intelligent Impact Force Modeling And Monitoring In Hislo Conditions For Maximizing Surface Mining Efficiency, Safety, And Health, Danish Ali

Doctoral Dissertations

"Shovel-truck systems are the most widely employed excavation and material handling systems for surface mining operations. During this process, a high-impact shovel loading operation (HISLO) produces large forces that cause extreme whole body vibrations (WBV) that can severely affect the safety and health of haul truck operators. Previously developed solutions have failed to produce satisfactory results as the vibrations at the truck operator seat still exceed the “Extremely Uncomfortable Limits”. This study was a novel effort in developing deep learning-based solution to the HISLO problem.

This research study developed a rigorous mathematical model and a 3D virtual simulation model to …

Adversarial Reconstruction Loss For Domain Generalization, Bekkouch Imad Eddine Ibrahim, Dragos Constantin Nicolae, Adil Khan, S. M. Ahsan Kazmi, Asad Masood Khattak, Bulat Ibragimov

All Works

The biggest fear when deploying machine learning models to the real world is their ability to handle the new data. This problem is significant especially in medicine, where models trained on rich high-quality data extracted from large hospitals do not scale to small regional hospitals. One of the clinical challenges addressed in this work is magnetic resonance image generalization for improved visualization and diagnosis of hip abnormalities such as femoroacetabular impingement and dysplasia. Domain Generalization (DG) is a field in machine learning that tries to solve the model’s dependency on the training data by leveraging many related but different data …

Pothole Detection Under Diverse Conditions Using Object Detection Model, Ibrahim Hassan Syed, Dympna O'Sullivan, Susan Mckeever

Datasets

One of the most important tasks in road maintenance is the detection of potholes. This process is usually done through manual visual inspection, where certified engineers assess recorded images of pavements acquired using cameras or professional road assessment vehicles. Machine learning techniques are now being applied to this problem, with models trained to automatically identify road conditions. However, approaching this real-world problem with machine learning techniques presents the classic problem of how to produce generalizable models. Images and videos may be captured in different illumination conditions, with different camera types, camera angles and resolutions. In this paper we present our …

Multi-Branch Gabor Wavelet Layers For Pedestrian Attribute Recognition, Imran N. Junejo

All Works

CCBYNCND Surveillance cameras are everywhere, keeping an eye on pedestrians as they navigate through a scene. With this context, our paper addresses the problem of pedestrian attribute recognition (PAR). This problem entails recognizing attributes such as age-group, clothing style, accessories, footwear style etc. This is a multi-label problem and challenging even for human observers. The problem has rightly attracted attention recently from the computer vision community. In this paper, we adopt trainable Gabor wavelets (TGW) layers and use it with a convolution neural network (CNN). Whereas other researchers are using fixed Gabor filters with the CNN, the proposed layers are …

A New Distributed Anomaly Detection Approach For Log Ids Management Based Ondeep Learning, Murat Koca, Muhammed Ali̇ Aydin, Ahmet Sertbaş, Abdül Hali̇m Zai̇m

Turkish Journal of Electrical Engineering and Computer Sciences

Today, with the rapid increase of data, the security of big data has become more important than ever for managers. However, traditional infrastructure systems cannot cope with increasingly big data that is created like an avalanche. In addition, as the existing database systems increase licensing costs per transaction, organizations using information technologies are shifting to free and open source solutions. For this reason, we propose an anomaly attack detection model on Apache Hadoop distributed file system (HDFS), which stands out in open source big data analytics, and Apache Spark, which stands out with its speed performance in analysis to reduce …

Just-In-Time Biomass Yield Estimation With Multi-Modal Data And Variable Patch Training Size, Patricia O'Byrne, Patrick Jackman Dr., Damon Dr. Berry Dr., Thomas Lee, Michael French, Robert J. Ross

Conference papers

The just-in-time estimation of farmland traits such as biomass yield can aid considerably in the optimisation of agricultural processes. Data in domains such as precision farming is however notoriously expensive to collect and deep learning driven modelling approaches need to maximise performance but also acknowledge this reality. In this paper we present a study in which a platform was deployed to collect data from a heterogeneous collection of sensor types including visual, NIR, and LiDAR sources to estimate key pastureland traits. In addition to introducing the study itself we address two key research questions. The first of these was the …

Affordance Learning For Visual-Semantic Perception, Chau Nguyen Duc Minh

Theses: Doctorates and Masters

Affordance Learning is linked to the study of interactions between robots and objects, including how robots perceive objects by scene understanding. This area has been popular in the Psychology, which has recently come to influence Computer Vision. In this way, Computer Vision has borrowed the concept of affordance from Psychology in order to develop Visual-Semantic recognition systems, and to develop the capabilities of robots to interact with objects, in particular. However, existing systems of Affordance Learning are still limited to detecting and segmenting object affordances, which is called Affordance Segmentation. Further, these systems are not designed to develop specific abilities …

Data: The Good, The Bad And The Ethical, John D. Kelleher, Filipe Cabral Pinto, Luis M. Cortesao

Articles

It is often the case with new technologies that it is very hard to predict their long-term impacts and as a result, although new technology may be beneficial in the short term, it can still cause problems in the longer term. This is what happened with oil by-products in different areas: the use of plastic as a disposable material did not take into account the hundreds of years necessary for its decomposition and its related long-term environmental damage. Data is said to be the new oil. The message to be conveyed is associated with its intrinsic value. But as in …

A Deep Learning Framework Supporting Model Ownership Protection And Traitor Tracing, Guowen Xu, Hongwei Li, Yuan Zhang, Xiaodong Lin, Robert H. Deng, Xuemin (Sherman) Shen

Research Collection School Of Computing and Information Systems

Cloud-based deep learning (DL) solutions have been widely used in applications ranging from image recognition to speech recognition. Meanwhile, as commercial software and services, such solutions have raised the need for intellectual property rights protection of the underlying DL models. Watermarking is the mainstream of existing solutions to address this concern, by primarily embedding pre-defined secrets in a model's training process. However, existing efforts almost exclusively focus on detecting whether a target model is pirated, without considering traitor tracing. In this paper, we present SecureMark_DL, which enables a model owner to embed a unique fingerprint for every customer within parameters …

Integrating Deep Learning And Augmented Reality To Enhance Situational Awareness In Firefighting Environments, Manish Bhattarai

Electrical and Computer Engineering ETDs

We present a new four-pronged approach to build firefighter's situational awareness for the first time in the literature. We construct a series of deep learning frameworks built on top of one another to enhance the safety, efficiency, and successful completion of rescue missions conducted by firefighters in emergency first response settings. First, we used a deep Convolutional Neural Network (CNN) system to classify and identify objects of interest from thermal imagery in real-time. Next, we extended this CNN framework for object detection, tracking, segmentation with a Mask RCNN framework, and scene description with a multimodal natural language processing(NLP) framework. Third, …

Meta-Rcnn: Meta Learning For Few-Shot Object Detection, Xiongwei Wu, Doyen Sahoo, Steven Hoi

Research Collection School Of Computing and Information Systems

Despite significant advances in deep learning based object detection in recent years, training effective detectors in a small data regime remains an open challenge. This is very important since labelling training data for object detection is often very expensive and time-consuming. In this paper, we investigate the problem of few-shot object detection, where a detector has access to only limited amounts of annotated data. Based on the meta-learning principle, we propose a new meta-learning framework for object detection named "Meta-RCNN", which learns the ability to perform few-shot detection via meta-learning. Specifically, Meta-RCNN learns an object detector in an episodic learning …

Tag: Automated Image Captioning, Nathan Funckes

McNair Scholars Manuscripts

Many websites remain non-ADA compliant, containing images which lack accompanying textual descriptions. This leaves sight-impaired individuals unable to fully enjoy the rich wonders of the web. To address this inequity, our research aims to create an autonomous system capable of generating semantically accurate descriptions of images. This problem involves two tasks: recognizing an image and linguistically describing it. Our solution uses state-of-the-art deep learning: employing a convolutional neural network that "learns" to understand images and extracts their salient features, and a recurrent neural network that learns to generate structured, coherent sentences. These two networks are merged to create a single …

A Study Of Information Bots And Knowledge Bots, Amartya Hatua

Dissertations

In this dissertation, a study of different aspects of information bots and knowledge bots is done. The research contributes to a better understanding of the various characteristics of information bots as well as the different patterns and factors responsible for the information diffusion in a social network. This research also shows how these factors can be used to predict information diffusion for a particular topic in a social network. The second part of the research is focused on strategies for improving the knowledge base of knowledge bots, where two different approaches are studied. In the first approach, knowledge is transferred …

Rethinking Pruning For Accelerating Deep Inference At The Edge, Dawei Gao, Xiaoxi He, Zimu Zhou, Yongxin Tong, Ke Xu, Lothar Thiele

Research Collection School Of Computing and Information Systems

There is a growing trend to deploy deep neural networks at the edge for high-accuracy, real-time data mining and user interaction. Applications such as speech recognition and language understanding often apply a deep neural network to encode an input sequence and then use a decoder to generate the output sequence. A promising technique to accelerate these applications on resource-constrained devices is network pruning, which compresses the size of the deep neural network without severe drop in inference accuracy. However, we observe that although existing network pruning algorithms prove effective to speed up the prior deep neural network, they lead to …

Deep Learning For Real-World Object Detection, Xiongwei Wu

Dissertations and Theses Collection (Open Access)

Despite achieving significant progresses, most existing detectors are designed to detect objects in academic contexts but consider little in real-world scenarios. In real-world applications, the scale variance of objects can be significantly higher than objects in academic contexts; In addition, existing methods are designed for achieving localization with relatively low precision, however more precise localization is demanded in real-world scenarios; Existing methods are optimized with huge amount of annotated data, but in certain real-world scenarios, only a few samples are available. In this dissertation, we aim to explore novel techniques to address these research challenges to make object detection algorithms …

Automating The Classification Of Mosquito Specimens Using Image Processing Techniques, Mona Minakshi

USF Tampa Graduate Theses and Dissertations

According to WHO (World Health Organization) reports, among all animals, mosquitoes are responsible for the most deaths worldwide. Mosquito borne diseases continue to pose grave dangers to global health. In 2015 alone, 214 million cases of malaria were registered worldwide. According to Centers for Disease Control and Prevention (CDC) report published in 2016, 62,500 suspected case of Zika were reported to the Puerto Rico Department of Health (PRDH) out of which 29,345 cases were found positive. The year 2019 was recorded as the worst for dengue in South East Asia. There are close to 4,500 species of mosquitoes (spread across …

Attacking Computer Vision Models Using Occlusion Analysis To Create Physically Robust Adversarial Images, Jacobsen Loh

Master's Theses

Self-driving cars rely on their sense of sight to function effectively in chaotic and uncontrolled environments. Thanks to recent developments in computer vision, specifically convolutional neural networks, autonomous vehicles have developed the ability to see at or above human-level capabilities, which in turn has allowed for rapid advances in self-driving cars. Unfortunately, much like humans being confused by simple optical illusions, convolutional neural networks are susceptible to simple adversarial inputs. As there is no overlap between the optical illusions that fool humans and the adversarial examples that threaten convolutional neural networks, little is understood as to why these adversarial examples …

Ai Quantification Of Language Puzzle To Language Learning Generalization, Harita Shroff

Master's Projects

Online language learning applications provide users multiple ways/games to learn a new language. Some of the ways include rearranging words in the foreign language sentences, filling in the blanks, providing flashcards, and many more. Primarily this research focused on quantifying the effectiveness of these games in learning a new language. Secondarily my goal for this project was to measure the effectiveness of exercises for transfer learning in machine translation. Currently, very little research has been done in this field except for the research conducted by the online platforms to provide assurance to their users [12]. Machine learning has been used …

Using Deep Learning And Linguistic Analysis To Predict Fake News Within Text, John Nguyen

Master's Projects

The spread of information about current events is a way for everybody in the world to learn and understand what is happening in the world. In essence, the news is an important and powerful tool that could be used by various groups of people to spread awareness and facts for the good of mankind. However, as information becomes easily and readily available for public access, the rise of deceptive news becomes an increasing concern. The reason is due to the fact that it will cause people to be misled and thus could affect the livelihood of themselves or others. The …

Deep Representation Learning On Giga-Pixel Whole Slide Images, Xinliang Zhu

Computer Science and Engineering Dissertations - Archive

I present my work towards solving the fundamental, challenging and valuable problem for automatically processing the giga-pixel level whole slide pathology images (WSIs): the representation of them. Specifically, I target on solving the combinations of three critical aspects of the problem: (1) it's not engineering feasible to directly fit them into existing convolutional neural networks because they are too large; (2) pre-trained parameters from other domains may not be effectively transferred to pathology images, and (3) both the image samples and annotations for those images are rarely available. To evaluate the effectiveness of the developed methods, I mainly focus on …

Towards Multi-Modal Data Classification, Henry Ng

UNLV Theses, Dissertations, Professional Papers, and Capstones

A feature fusion multi-modal neural network (MMN) is a network that combines different modalities at the feature level to perform a specific task. In this paper, we study the problem of training the fusion procedure for MMN. A recent study has found that training a multi-modal network that incorporates late fusion produces a network that has not learned the proper parameters for feature extraction. These late fusion models perform very well during training but fall short to its single modality counterpart when testing. We hypothesize that jointly trained MMN have weight space that is too large for effective training. To …

Deep Learning Framework Of Vehicle Detection And Tracking System, Rui Zhang

Other Student Works

In the last semester, I designed a system to detect and track vehicle system on the highway. The system is based on the public deep learning framework and utilize pre-trained model to implement the functions of this system. In this paper, I will use my own framework to implement this system. This try will help us better understand the details of deep learning framework. I will public the code of deep learning framework and make sure everyone can modify it.

This semester will focus on the researching of Naïve Convolutional Neural Networks. Neural networks are commonlyused for the analysis of …

Book Genre Classification By Its Cover Using A Multi-View Learning Approach, Chandra Shakhar Kundu

Masters Theses & Specialist Projects

An interesting topic in the visual analysis is to determine the genre of a book by its cover. The book cover is the very first communication to the reader which shapes the reader’s expectation about the type of the book. Each book cover is carefully designed by the cover designers and typographers to convey the visual representation of its content. In this study, we explore several different deep learning approaches for predicting the genre from the cover image alone, such as MobileNet V1, MobileNet V2, ResNet50, Inception V2. Moreover, we add an extra modality by extracting text from the cover …

Improving Ocr Accuracy Of Damaged Pictures With Generative Adversarial Networks, Pu Du

LSU Master's Theses

In this thesis, we focus on resolving the inpainting problem and improving Optical Character Recognition (OCR) accuracy of damaged text images at character level. We present a Generative Adversarial Network (GAN)-based model conditioned on class labels for image inpainting. This model is a deep convolutional neural network with encoder-decoder style architecture which can process images with holes at random locations. Experiments on the character images dataset demonstrate that our proposed model generates promising inpainting results and significantly improve OCR accuracy by reconstructing missing parts of damaged character images.

Robust Neural Machine Translation, Abdul Rafae Khan

Dissertations, Theses, and Capstone Projects

This thesis aims for general robust Neural Machine Translation (NMT) that is agnostic to the test domain. NMT has achieved high quality on benchmarks with closed datasets such as WMT and NIST but can fail when the translation input contains noise due to, for example, mismatched domains or spelling errors. The standard solution is to apply domain adaptation or data augmentation to build a domain-dependent system. However, in real life, the input noise varies in a wide range of domains and types, which is unknown in the training phase. This thesis introduces five general approaches to improve NMT accuracy and …

Computational Model For Neural Architecture Search, Ram Deepak Gottapu

Doctoral Dissertations

"A long-standing goal in Deep Learning (DL) research is to design efficient architectures for a given dataset that are both accurate and computationally inexpensive. At present, designing deep learning architectures for a real-world application requires both human expertise and considerable effort as they are either handcrafted by careful experimentation or modified from a handful of existing models. This method is inefficient as the process of architecture design is highly time-consuming and computationally expensive.

The research presents an approach to automate the process of deep learning architecture design through a modeling procedure. In particular, it first introduces a framework that treats …