Digital Commons Network^™

Visual Odometry Using Convolutional Neural Networks, Alec Graves, Steffen Lim, Thomas Fagan, Kevin Mcfall Phd.

The Kennesaw Journal of Undergraduate Research

Visual odometry is the process of tracking an agent's motion over time using a visual sensor. The visual odometry problem has only been recently solved using traditional, non-machine learning techniques. Despite the success of neural networks at many related problems such as object recognition, feature detection, and optical flow, visual odometry still has not been solved with a deep learning technique. This paper attempts to implement several Convolutional Neural Networks to solve the visual odometry problem and compare slight variations in data preprocessing. The work presented is a step toward reaching a legitimate neural network solution.

How To Best Apply Neural Networks In Geosciences: Towards Optimal "Averaging" In Dropout Training, Afshin Gholamy, Justin Parra, Vladik Kreinovich, Olac Fuentes, Elizabeth Y. Anthony

Departmental Technical Reports (CS)

The main objectives of geosciences is to find the current state of the Earth -- i.e., solve the corresponding inverse problems -- and to use this knowledge for predicting the future events, such as earthquakes and volcanic eruptions. In both inverse and prediction problems, often, machine learning techniques are very efficient, and at present, the most efficient machine learning technique is deep neural training. To speed up this training, the current learning algorithms use dropout techniques: they train several sub-networks on different portions of data, and then "average" the results. A natural idea is to use arithmetic mean for this …

Online Learning With Nonlinear Models, Doyen Sahoo

Dissertations and Theses Collection (Open Access)

Recent years have witnessed the success of two broad categories of machine learning algorithms: (i) Online Learning; and (ii) Learning with nonlinear models. Typical machine learning algorithms assume that the entire data is available prior to the training task. This is often not the case in the real world, where data often arrives sequentially in a stream, or is too large to be stored in memory. To address these challenges, Online Learning techniques evolved as a promising solution to having highly scalable and efficient learning methodologies which could learn from data arriving sequentially. Next, as the real world data exhibited …

A Deep Learníng-Based Data Minimization Algorithm For Big Genomics Data In Support Of Lot And Secure Smart Health Services, Mohammed Aledhari

Dissertations

In the age of Big Genomics Data, institutes such as the National Human Genome Research Institute (NHGRI),1000-Genomes project, and the international cancer sequencing consortium are faced with the challenge of sharing large volumes of data between internationallydispersed sample collectors, data analyzers, and researchers, a process that up until now has been plagued by unreliable transfers and slow connection speeds. These occur due to the inherent throughput bottlenecks of traditional transfer technologies. One suggested solution is using the cloud as an infrastructure to solve the store and analysis challenges. However, the transfer and share of the genomics datasets between biological laboratories …

Deep-Learned Generative Representations Of 3d Shape Families, Haibin Huang

Doctoral Dissertations

Digital representations of 3D shapes are becoming increasingly useful in several emerging applications, such as 3D printing, virtual reality and augmented reality. However, traditional modeling softwares require users to have extensive modeling experience, artistic skills and training to handle their complex interfaces and perform the necessary low-level geometric manipulation commands. Thus, there is an emerging need for computer algorithms that help novice and casual users to quickly and easily generate 3D content. In this work, I will present deep learning algorithms that are capable of automatically inferring parametric representations of shape families, which can be used to generate new 3D …

Deep Energy-Based Models For Structured Prediction, David Belanger

Doctoral Dissertations

We introduce structured prediction energy networks (SPENs), a flexible frame- work for structured prediction. A deep architecture is used to define an energy func- tion over candidate outputs and predictions are produced by gradient-based energy minimization. This deep energy captures dependencies between labels that would lead to intractable graphical models, and allows us to automatically discover discrim- inative features of the structured output. Furthermore, practitioners can explore a wide variety of energy function architectures without having to hand-design predic- tion and learning methods for each model. This is because all of our prediction and learning methods interact with the energy …

Intent Recognition In Smart Living Through Deep Recurrent Neural Networks, Xiang Zhang, Lina Yao, Chaoran Huang, Quan Z. Sheng, Xianzhi Wang

Research Collection School Of Computing and Information Systems

Electroencephalography (EEG) signal based intent recognition has recently attracted much attention in both academia and industries, due to helping the elderly or motor-disabled people controlling smart devices to communicate with outer world. However, the utilization of EEG signals is challenged by low accuracy, arduous and time-consuming feature extraction. This paper proposes a 7-layer deep learning model to classify raw EEG signals with the aim of recognizing subjects’ intents, to avoid the time consumed in pre-processing and feature extraction. The hyper-parameters are selected by an Orthogonal Array experiment method for efficiency. Our model is applied to an open EEG dataset provided …

Hierarchical Fusion Based Deep Learning Framework For Lung Nodule Classification, Kazim Sekeroglu

LSU Doctoral Dissertations

Lung cancer is the leading cancer type that causes the mortality in both men and women. Computer aided detection (CAD) and diagnosis systems can play a very important role for helping the physicians in cancer treatments. This dissertation proposes a CAD framework that utilizes a hierarchical fusion based deep learning model for detection of nodules from the stacks of 2D images. In the proposed hierarchical approach, a decision is made at each level individually employing the decisions from the previous level. Further, individual decisions are computed for several perspectives of a volume of interest (VOI). This study explores three different …

Exploring The Internal Statistics: Single Image Super-Resolution, Completion And Captioning, Yang Xian

Dissertations, Theses, and Capstone Projects

Image enhancement has drawn increasingly attention in improving image quality or interpretability. It aims to modify images to achieve a better perception for human visual system or a more suitable representation for further analysis in a variety of applications such as medical imaging, remote sensing, and video surveillance. Based on different attributes of the given input images, enhancement tasks vary, e.g., noise removal, deblurring, resolution enhancement, prediction of missing pixels, etc. The latter two are usually referred to as image super-resolution and image inpainting (or completion).

Image super-resolution and completion are numerically ill-posed problems. Multi-frame-based approaches make use of the …

Automated Breast Cancer Diagnosis Using Deep Learning And Region Of Interest Detection (Bc-Droid), Richard Platania, Jian Zhang, Shayan Shams, Kisung Lee, Seungwon Yang, Seung Jong Park

Computer Science Faculty Research & Creative Works

Detection of suspicious regions in mammogram images and the subsequent diagnosis of these regions remains a challenging problem in the medical world. There still exists an alarming rate of misdiagnosis of breast cancer. This results in both over treatment through incorrect positive diagnosis of cancer and under treatment through overlooked cancerous masses. Convolutional neural networks have shown strong applicability to various image datasets, enabling detailed features to be learned from the data and, as a result, the ability to classify these images at extremely low error rates. In order to overcome the difficulty in diagnosing breast cancer from mammogram images, …

Deepfacade: A Deep Learning Approach To Facade Parsing, Hantang Liu, Jialiang Zhang, Jianke Zhu, Steven C. H. Hoi

Research Collection School Of Computing and Information Systems

The parsing of building facades is a key component to the problem of 3D street scenes reconstruction, which is long desired in computer vision. In this paper, we propose a deep learning based method for segmenting a facade into semantic categories. Man-made structures often present the characteristic of symmetry. Based on this observation, we propose a symmetric regularizer for training the neural network. Our proposed method can make use of both the power of deep neural networks and the structure of man-made architectures. We also propose a method to refine the segmentation results using bounding boxes generated by the Region …

Formresnet: Formatted Residual Learning For Image Restoration, Jianbo Jiao, Wei-Chih Tu, Shengfeng He

Research Collection School Of Computing and Information Systems

In this paper, we propose a deep CNN to tackle the image restoration problem by learning the structured residual. Previous deep learning based methods directly learn the mapping from corrupted images to clean images, and may suffer from the gradient exploding/vanishing problems of deep neural networks. We propose to address the image restoration problem by learning the structured details and recovering the latent clean image together, from the shared information between the corrupted image and the latent image. In addition, instead of learning the pure difference (corruption), we propose to add a 'residual formatting layer' to format the residual to …

An Ensemble Deep Convolutional Neural Network Model With Improved D-S Evidence Fusion For Bearing Fault Diagnosis, Shaobo Li, Guoka Liu, Xianghong Tang, Jianguang Lu, Jianjun Hu

Faculty Publications

Intelligent machine health monitoring and fault diagnosis are becoming increasingly important for modern manufacturing industries. Current fault diagnosis approaches mostly depend on expert-designed features for building prediction models. In this paper, we proposed IDSCNN, a novel bearing fault diagnosis algorithm based on ensemble deep convolutional neural networks and an improved Dempster–Shafer theory based evidence fusion. The convolutional neural networks take the root mean square (RMS) maps from the FFT (Fast Fourier Transformation) features of the vibration signals from two sensors as inputs. The improved D-S evidence theory is implemented via distance matrix from evidences and modified Gini Index. Extensive evaluations …

Evaluation Of Deep Learning Frameworks Over Different Hpc Architectures, Shayan Shams, Richard Platania, Kisung Lee, Seung Jong Park

Computer Science Faculty Research & Creative Works

Recent advances in deep learning have enabled researchers across many disciplines to uncover new insights about large datasets. Deep neural networks have shown applicability to image, time-series, textual, and other data, all of which are available in a plethora of research fields. However, their computational complexity and large memory overhead requires advanced software and hardware technologies to train neural networks in a reasonable amount of time. To make this possible, there has been an influx in development of deep learning software that aim to leverage advanced hardware resources. In order to better understand the performance implications of deep learning frameworks …

Speech Based Machine Learning Models For Emotional State Recognition And Ptsd Detection, Debrup Banerjee

Electrical & Computer Engineering Theses & Dissertations

Recognition of emotional state and diagnosis of trauma related illnesses such as posttraumatic stress disorder (PTSD) using speech signals have been active research topics over the past decade. A typical emotion recognition system consists of three components: speech segmentation, feature extraction and emotion identification. Various speech features have been developed for emotional state recognition which can be divided into three categories, namely, excitation, vocal tract and prosodic. However, the capabilities of different feature categories and advanced machine learning techniques have not been fully explored for emotion recognition and PTSD diagnosis. For PTSD assessment, clinical diagnosis through structured interviews is a …

Back To The Future: Logic And Machine Learning, Simon Dobnik, John D. Kelleher

Conference papers

In this paper we argue that since the beginning of the natural language processing or computational linguistics there has been a strong connection between logic and machine learning. First of all, there is something logical about language or linguistic about logic. Secondly, we argue that rather than distinguishing between logic and machine learning, a more useful distinction is between top-down approaches and data-driven approaches. Examining some recent approaches in deep learning we argue that they incorporate both properties and this is the reason for their very successful adoption to solve several problems within language technology.

Tackling The Interleaving Problem In Activity Discovery, Eoin Rogers, Robert J. Ross, John D. Kelleher

Conference papers

Activity discovery (AD) is the unsupervised process of discovering activities in data produced from streaming sensor networks that are recording the actions of human subjects. One major challenge for AD systems is interleaving, the tendency for people to carry out multiple activities at a time a parallel. Following on from our previous work, we continue to investigate AD in interleaved datasets, with a view towards progressing the state-of-the-art for AD.

Deepmon: Mobile Gpu-Based Deep Learning Framework For Continuous Vision Applications, Nguyen Loc Huynh, Youngki Lee, Rajesh Krishna Balan

Research Collection School Of Computing and Information Systems

The rapid emergence of head-mounted devices such as the Microsoft Holo-lens enables a wide variety of continuous vision applications. Such applications often adopt deep-learning algorithms such as CNN and RNN to extract rich contextual information from the first-person-view video streams. Despite the high accuracy, use of deep learning algorithms in mobile devices raises critical challenges, i.e., high processing latency and power consumption. In this paper, we propose DeepMon, a mobile deep learning inference system to run a variety of deep learning inferences purely on a mobile device in a fast and energy-efficient manner. For this, we designed a suite of …

Demo: Deepmon - Building Mobile Gpu Deep Learning Models For Continuous Vision Applications, Loc Nguyen Huynh, Rajesh Krishna Balan, Youngki Lee

Research Collection School Of Computing and Information Systems

Deep learning has revolutionized vision sensing applications in terms of accuracy comparing to other techniques. Its breakthrough comes from the ability to extract complex high level features directly from sensor data. However, deep learning models are still yet to be natively supported on mobile devices due to high computational requirements. In this paper, we present DeepMon, a next generation of DeepSense [1] framework, to enable deep learning models on conventional mobile devices (e.g. Samsung Galaxy S7) for continuous vision sensing applications. Firstly, Deep-Mon exploits similarity between consecutive video frames for intermediate data caching within models to enhance inference latency. Secondly, …

Aspect Discovery From Product Reviews, Ying Ding

Dissertations and Theses Collection

With the rapid development of online shopping sites and social media, product reviews are accumulating. These reviews contain information that is valuable to both businesses and customers. To businesses, companies can easily get a large number of feedback of their products, which is difficult to achieve by doing customer survey in the traditional way. To customers, they can know the products they are interested in better by reading reviews, which may be uneasy without online reviews. However, the accumulation has caused consuming all reviews impossible. It is necessary to develop automated techniques to efficiently process them. One of the most …

Viewability Prediction For Display Advertising, Chong Wang

Dissertations

As a massive industry, display advertising delivers advertisers’ marketing messages to attract customers through graphic banners on webpages. Display advertising is also the most essential revenue source of online publishers. Currently, advertisers are charged by user response or ad serving. However, recent studies show that users barely click or convert display ads. Moreover, about half of the ads are actually never seen by users. In this case, advertisers cannot enhance their brand awareness and increase return on investment. Publishers also lose much revenue. Therefore, the ad pricing standards are shifting to a new model: ad impressions are paid if they …

Investigation Of New Learning Methods For Visual Recognition, Qingfeng Liu

Dissertations

Visual recognition is one of the most difficult and prevailing problems in computer vision and pattern recognition due to the challenges in understanding the semantics and contents of digital images. Two major components of a visual recognition system are discriminatory feature representation and efficient and accurate pattern classification. This dissertation therefore focuses on developing new learning methods for visual recognition.

Based on the conventional sparse representation, which shows its robustness for visual recognition problems, a series of new methods is proposed. Specifically, first, a new locally linear K nearest neighbor method, or LLK method, is presented. The LLK method derives …

A Compare-Aggregate Model For Matching Text Sequences, Shuohang Wang, Jing Jiang

Research Collection School Of Computing and Information Systems

Many NLP tasks including machine comprehension, answer selection and text entailment require the comparison between sequences. Matching the important units between sequences is a key to solve these problems. In this paper, we present a general "compare-aggregate" framework that performs word-level matching followed by aggregation using Convolutional Neural Networks. We particularly focus on the different comparison functions we can use to match two vectors. We use four different datasets to evaluate the model. We find that some simple comparison functions based on element-wise operations can work better than standard neural network and neural tensor network.

Ai Education: Deep Neural Network Learning Resources, Todd W. Neller

Computer Science Faculty Publications

In this column, we focus on resources for learning and teaching deep neural network learning. Many exciting advances have been made in this area of late, and so many resources have become available online that the flood of relevant concepts and techniques can be overwhelming. Here, we hope to provide a sampling of high-quality resources to guide the newcomer into this booming field. [excerpt]

Deep Learning Method Vs. Hand-Crafted Features For Lung Cancer Diagnosis And Breast Cancer Risk Analysis, Wenqing Sun

Open Access Theses & Dissertations

Breast cancer and lung cancer are two major leading causes of cancer deaths, and researchers have been developing computer aided diagnosis (CAD) system to automatically diagnose them for decades. In recent studies, we found that the techniques in CAD system can also be used for breast cancer risk analysis, like feature design and machine learning. Also we noticed that with the development of deep learning methods, the performance of CAD system can be improved by using computer automatically generated features. To explore these possibilities, we conducted a series of studies: the first two studies focused on transferring the original CAD …

Review Of Trends In Health Social Media Analysis, Liliya Akhtyamova, Mikhail Alexandrov, John Cardiff

Conference Papers

This paper surveys recent publications (2008-2017) on using social media data to study public health. The survey describes the main topics being discussed in forums and presents short information about methods and tools used for analysis health social media. We put especial attention on adverse drug reaction detection problem (ADR).

Deep Models For Engagement Assessment With Scarce Label Information, Feng Li, Guangfan Zhang, Wei Wang, Roger Xu, Tom Schnell, Jonathan Wen, Frederic Mckenzie, Jiang Li

Electrical & Computer Engineering Faculty Publications

Task engagement is defined as loadings on energetic arousal (affect), task motivation, and concentration (cognition) [1]. It is usually challenging and expensive to label cognitive state data, and traditional computational models trained with limited label information for engagement assessment do not perform well because of overfitting. In this paper, we proposed two deep models (i.e., a deep classifier and a deep autoencoder) for engagement assessment with scarce label information. We recruited 15 pilots to conduct a 4-h flight simulation from Seattle to Chicago and recorded their electroencephalograph (EEG) signals during the simulation. Experts carefully examined the EEG signals and labeled …

An Alternative Approach To Training Sequence-To-Sequence Model For Machine Translation, Vivek Sah

Honors Theses

Machine translation is a widely researched topic in the field of Natural Language Processing and most recently, neural network models have been shown to be very effective at this task. The model, called sequence-to-sequence model, learns to map an input sequence in one language to a vector of fixed dimensionality and then map that vector to an output sequence in another language without any human intervention provided that there is enough training data. Focusing on English-French translation, in this paper, I present a way to simplify the learning process by replacing English input sentences by word-by-word translation of those sentences. …

Symbolic And Deep Learning Based Data Representation Methods For Activity Recognition And Image Understanding At Pixel Level, Manohar Karki

LSU Doctoral Dissertations

Efficient representation of large amount of data particularly images and video helps in the analysis, processing and overall understanding of the data. In this work, we present two frameworks that encapsulate the information present in such data. At first, we present an automated symbolic framework to recognize particular activities in real time from videos. The framework uses regular expressions for symbolically representing (possibly infinite) sets of motion characteristics obtained from a video. It is a uniform framework that handles trajectory-based and periodic articulated activities and provides polynomial time graph algorithms for fast recognition. The regular expressions representing motion characteristics can …