Artificial Intelligence and Robotics Commons^™

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 …

Unitary And Symmetric Structure In Deep Neural Networks, Kehelwala Dewage Gayan Maduranga

Theses and Dissertations--Mathematics

Recurrent neural networks (RNNs) have been successfully used on a wide range of sequential data problems. A well-known difficulty in using RNNs is the vanishing or exploding gradient problem. Recently, there have been several different RNN architectures that try to mitigate this issue by maintaining an orthogonal or unitary recurrent weight matrix. One such architecture is the scaled Cayley orthogonal recurrent neural network (scoRNN), which parameterizes the orthogonal recurrent weight matrix through a scaled Cayley transform. This parametrization contains a diagonal scaling matrix consisting of positive or negative one entries that can not be optimized by gradient descent. Thus the …

Estimating Free-Flow Speed With Lidar And Overhead Imagery, Armin Hadzic

Theses and Dissertations--Computer Science

Understanding free-flow speed is fundamental to transportation engineering in order to improve traffic flow, control, and planning. The free-flow speed of a road segment is the average speed of automobiles unaffected by traffic congestion or delay. Collecting speed data across a state is both expensive and time consuming. Some approaches have been presented to estimate speed using geometric road features for certain types of roads in limited environments. However, estimating speed at state scale for varying landscapes, environments, and road qualities has been relegated to manual engineering and expensive sensor networks. This thesis proposes an automated approach for estimating free-flow …

Exploration And Implementation Of Neural Ordinary Differential Equations, Long Huu Nguyen, Andy Malinsky

Capstone Showcase

Neural ordinary differential equations (ODEs) have recently emerged as a novel ap- proach to deep learning, leveraging the knowledge of two previously separate domains, neural networks and differential equations. In this paper, we first examine the back- ground and lay the foundation for traditional artificial neural networks. We then present neural ODEs from a rigorous mathematical perspective, and explore their advantages and trade-offs compared to traditional neural nets.

Catgame: A Tool For Problem Solving In Complex Dynamic Systems Using Game Theoretic Knowledge Distribution In Cultural Algorithms, And Its Application (Catneuro) To The Deep Learning Of Game Controller, Faisal Waris

Wayne State University Dissertations

Cultural Algorithms (CA) are knowledge-intensive, population-based stochastic optimization methods that are modeled after human cultures and are suited to solving problems in complex environments. The CA Belief Space stores knowledge harvested from prior generations and re-distributes it to future generations via a knowledge distribution (KD) mechanism. Each of the population individuals is then guided through the search space via the associated knowledge. Previously, CA implementations have used only competitive KD mechanisms that have performed well for problems embedded in static environments. Relatively recently, CA research has evolved to encompass dynamic problem environments. Given increasing environmental complexity, a natural question arises …

Cheat Detection Using Machine Learning Within Counter-Strike: Global Offensive, Harry Dunham

Senior Independent Study Theses

Deep learning is becoming a steadfast means of solving complex problems that do not have a single concrete or simple solution. One complex problem that fits this description and that has also begun to appear at the forefront of society is cheating, specifically within video games. Therefore, this paper presents a means of developing a deep learning framework that successfully identifies cheaters within the video game CounterStrike: Global Offensive. This approach yields predictive accuracy metrics that range between 80-90% depending on the exact neural network architecture that is employed. This approach is easily scalable and applicable to all types of …

Representation Learning With Adversarial Latent Autoencoders, Stanislav Pidhorskyi M.S.

Graduate Theses, Dissertations, and Problem Reports (ETD)

A large number of deep learning methods applied to computer vision problems require encoder-decoder maps. These methods include, but are not limited to, self-representation learning, generalization, few-shot learning, and novelty detection. Encoder-decoder maps are also useful for photo manipulation, photo editing, superresolution, etc. Encoder-decoder maps are typically learned using autoencoder networks.
Traditionally, autoencoder reciprocity is achieved in the image-space using pixel-wise
similarity loss, which has a widely known flaw of producing non-realistic reconstructions. This flaw is typical for the Variational Autoencoder (VAE) family and is not only limited to pixel-wise similarity losses, but is common to all methods relying upon …

Ordinal Hyperplane Loss, Bob Vanderheyden

Doctor of Data Science and Analytics Dissertations

This research presents the development of a new framework for analyzing ordered class data, commonly called “ordinal class” data. The focus of the work is the development of classifiers (predictive models) that predict classes from available data. Ratings scales, medical classification scales, socio-economic scales, meaningful groupings of continuous data, facial emotional intensity and facial age estimation are examples of ordinal data for which data scientists may be asked to develop predictive classifiers. It is possible to treat ordinal classification like any other classification problem that has more than two classes. Specifying a model with this strategy does not fully utilize …

Similarity-Based Chained Transfer Learning For Energy Forecasting With Big Data, Yifang Tian, Ljubisa Sehovac, Katarina Grolinger

Electrical and Computer Engineering Publications

Smart meter popularity has resulted in the ability to collect big energy data and has created opportunities for large-scale energy forecasting. Machine Learning (ML) techniques commonly used for forecasting, such as neural networks, involve computationally intensive training typically with data from a single building or a single aggregated load to predict future consumption for that same building or aggregated load. With hundreds of thousands of meters, it becomes impractical or even infeasible to individually train a model for each meter. Consequently, this paper proposes Similarity-Based Chained Transfer Learning (SBCTL), an approach for building neural network-based models for many meters by …

Pristine Sentence Translation: A New Approach To A Timeless Problem, Meenu Ahluwalia, Brian Coari, Ben Brock

SMU Data Science Review

Abstract.

Pristine Sentence Translation (PST) is a new approach to language translation based upon sentence-level granularity. Traditional translation approaches, including those utilizing advanced machine learning or neural network-based approaches, translate on a word-by-word or phrase-by-phrase basis; thereby, potentially missing the context or meaning of the complete sentence. Instead of these piecewise translations, PST utilizes deep learning and predictive modeling techniques to translate complete sentences from their source language into their target language. With these approaches we were able to translate sentences that closely conveyed the meaning of the original sentences. Our results demonstrated that PST’s method of translating an entire …

Texture-Based Deep Neural Network For Histopathology Cancer Whole Slide Image (Wsi) Classification, Nelson Zange Tsaku

Master of Science in Computer Science Theses

Automatic histopathological Whole Slide Image (WSI) analysis for cancer classification has been highlighted along with the advancements in microscopic imaging techniques. However, manual examination and diagnosis with WSIs is time-consuming and tiresome. Recently, deep convolutional neural networks have succeeded in histopathological image analysis. In this paper, we propose a novel cancer texture-based deep neural network (CAT-Net) that learns scalable texture features from histopathological WSIs. The innovation of CAT-Net is twofold: (1) capturing invariant spatial patterns by dilated convolutional layers and (2) Reducing model complexity while improving performance. Moreover, CAT-Net can provide discriminative texture patterns formed on cancerous regions of histopathological …

Towards Robust Resnet: A Small Step But A Giant Leap, Jingfeng Zhang, Bo Han, Laura Wynter, Bryan Kian Hsiang Low, Mohan Kankanhalli

Research Collection School Of Computing and Information Systems

This paper presents a simple yet principled approach to boosting the robustness of the residual network (ResNet) that is motivated by a dynamical systems perspective. Namely, a deep neural network can be interpreted using a partial differential equation, which naturally inspires us to characterize ResNet based on an explicit Euler method. This consequently allows us to exploit the step factor h in the Euler method to control the robustness of ResNet in both its training and generalization. In particular, we prove that a small step factor h can benefit its training and generalization robustness during backpropagation and forward propagation, respectively. …

A Study On Large-Scale Deep Learning In Bioinformatics And Biomedical Applications, Shayan Shams

LSU Doctoral Dissertations

Recent advances in Artificial Intelligence and deep learning have provided researchers in various fields insights into the analysis of multiple datasets. These applications include image analysis, text analysis, and many more. However, the effectiveness of deep learning in some areas, such as biomedical imaging and genomic research, has been overshadowed by the variance in the types and complexity of data. This is in addition to the expensive labeling process and the limited size of datasets in these fields. These challenges require advanced deep learning models capable of learning from a small dataset and also from a small number of labeled …

Forecasting Building Energy Consumption With Deep Learning: A Sequence To Sequence Approach, Ljubisa Sehovac, Cornelius Nesen, Katarina Grolinger

Electrical and Computer Engineering Publications

Energy Consumption has been continuously increasing due to the rapid expansion of high-density cities, and growth in the industrial and commercial sectors. To reduce the negative impact on the environment and improve sustainability, it is crucial to efficiently manage energy consumption. Internet of Things (IoT) devices, including widely used smart meters, have created possibilities for energy monitoring as well as for sensor based energy forecasting. Machine learning algorithms commonly used for energy forecasting such as feedforward neural networks are not well-suited for interpreting the time dimensionality of a signal. Consequently, this paper uses Recurrent Neural Networks (RNN) to capture time …

Deep Learning Based Real Time Devanagari Character Recognition, Aseem Chhabra

Master's Projects

The revolutionization of the technology behind optical character recognition (OCR) has helped it to become one of those technologies that have found plenty of uses in the entire industrial space. Today, the OCR is available for several languages and have the capability to recognize the characters in real time, but there are some languages for which this technology has not developed much. All these advancements have been possible because of the introduction of concepts like artificial intelligence and deep learning. Deep Neural Networks have proven to be the best choice when it comes to a task involving recognition. There are …

Learning To Play The Trading Game, Neeraj Kulkarni

Master's Projects

Can we train a stock trading bot that can take decisions in high-entropy envi- ronments like stock markets to generate profits based on some optimal policy? Can we further extend this learning for any general trading problem? Quantitative Al- gorithms are responsible for more than 75% of the stock trading around the world. Creating a stock market prediction model is comparatively easy. But creating a prof- itable prediction model is still considered as a challenging task in the field of machine learning and deep learning due to the unpredictability of the financial markets. Us- ing biologically inspired computing techniques of …

Pose Estimation And Action Recognition In Sports And Fitness, Parth Vyas

Master's Projects

The emergence of large datasets and major improvements in Deep Learning has lead to many real-world applications. These applications have been focused on automotive markets, mobile markets, stock markets, and the healthcare market. Although Deep Learning has strong foundations across many areas, the few applications in Sports, Fitness, or even Injury Rehabilitation could benefit greatly from it. For example, if you are performing a workout and you need to evaluate your form, but do not have access or resources for an instructor to evaluate your form, it would be great to have an Artificial Intelligent agent provide real time feedback …

Detecting Cars In A Parking Lot Using Deep Learning, Samuel Ordonia

Master's Projects

Detection of cars in a parking lot with deep learning involves locating all objects of interest in a parking lot image and classifying the contents of all bounding boxes as cars. Because of the variety of shape, color, contrast, pose, and occlusion, a deep neural net was chosen to encompass all the significant features required by the detector to differentiate cars from not cars. In this project, car detection was accomplished with a convolutional neural net (CNN) based on the You Only Look Once (YOLO) model architectures. An application was built to train and validate a car detection CNN as …

Self-Driving Cars: Evaluation Of Deep Learning Techniques For Object Detection In Different Driving Conditions, Ramesh Simhambhatla, Kevin Okiah, Shravan Kuchkula, Robert Slater

SMU Data Science Review

Deep Learning has revolutionized Computer Vision, and it is the core technology behind capabilities of a self-driving car. Convolutional Neural Networks (CNNs) are at the heart of this deep learning revolution for improving the task of object detection. A number of successful object detection systems have been proposed in recent years that are based on CNNs. In this paper, an empirical evaluation of three recent meta-architectures: SSD (Single Shot multi-box Detector), R-CNN (Region-based CNN) and R-FCN (Region-based Fully Convolutional Networks) was conducted to measure how fast and accurate they are in identifying objects on the road, such as vehicles, pedestrians, …

Leveraging Natural Language Processing Applications And Microblogging Platform For Increased Transparency In Crisis Areas, Ernesto Carrera-Ruvalcaba, Johnson Ekedum, Austin Hancock, Ben Brock

SMU Data Science Review

Through microblogging applications, such as Twitter, people actively document their lives even in times of natural disasters such as hurricanes and earthquakes. While first responders and crisis-teams are able to help people who call 911, or arrive at a designated shelter, there are vast amounts of information being exchanged online via Twitter that provide real-time, location-based alerts that are going unnoticed. To effectively use this information, the Tweets must be verified for authenticity and categorized to ensure that the proper authorities can be alerted. In this paper, we create a Crisis Message Corpus from geotagged Tweets occurring during 7 hurricanes …

The Challenge Of Collaborative Iot-Based Inferencing In Adversarial Settings, Archan Misra, Dulanga Kaveesha Weerakoon Weerakoon Mudiyanselage, Kasthuri Jayarajah

Research Collection School Of Computing and Information Systems

In many practical environments, resource-constrained IoT nodes are deployed with varying degrees of redundancy/overlap--i.e., their data streams possess significant spatiotemporal correlation. We posit that collaborative inferencing, whereby individual nodes adjust their inferencing pipelines to incorporate such correlated observations from other nodes, can improve both inferencing accuracy and performance metrics (such as latency and energy overheads). However, such collaborative models are vulnerable to adversarial behavior by one or more nodes, and thus require mechanisms that identify and inoculate against such malicious behavior. We use a dataset of 8 outdoor cameras to (a) demonstrate that such collaborative inferencing can improve people counting …

Dependable Machine Intelligence At The Tactical Edge, Archan Misra, Kasthuri Jayarajah, Dulanga Kaveesha Weerakoon Weerakoon Mudiyanselage, Randy Tandriansyah Daratan, Shuochao Yao, Tarek Abdelzaher

Research Collection School Of Computing and Information Systems

The paper describes a vision for dependable application of machine learning-based inferencing on resource-constrained edge devices. The high computational overhead of sophisticated deep learning learning techniques imposes a prohibitive overhead, both in terms of energy consumption and sustainable processing throughput, on such resource-constrained edge devices (e.g., audio or video sensors). To overcome these limitations, we propose a ``cognitive edge" paradigm, whereby (a) an edge device first autonomously uses statistical analysis to identify potential collaborative IoT nodes, and (b) the IoT nodes then perform real-time sharing of various intermediate state to improve their individual execution of machine intelligence tasks. We provide …

Neural Machine Translation, Quinn M. Lanners, Thomas Laurent

Honors Thesis

Neural Machine Translation is the primary algorithm used in industry to perform machine translation. This state-of-the-art algorithm is an application of deep learning in which massive datasets of translated sentences are used to train a model capable of translating between any two languages. The architecture behind neural machine translation is composed of two recurrent neural networks used together in tandem to create an Encoder Decoder structure. Attention mechanisms have recently been developed to further increase the accuracy of these models. In this senior thesis, the various parts of Neural Machine Translation are explored towards the eventual creation of a tutorial …

Deep Learning Based Medical Image Analysis With Limited Data, Jiaxing Tan

Dissertations, Theses, and Capstone Projects

Deep Learning Methods have shown its great effort in the area of Computer Vision. However, when solving the problems of medical imaging, deep learning’s power is confined by limited data available. We present a series of novel methodologies for solving medical imaging analysis problems with limited Computed tomography (CT) scans available. Our method, based on deep learning, with different strategies, including using Generative Adversar- ial Networks, two-stage training, infusing the expert knowledge, voting based or converting to other space, solves the data set limitation issue for the cur- rent medical imaging problems, specifically cancer detection and diagnosis, and shows very …

Improving Vix Futures Forecasts Using Machine Learning Methods, James Hosker, Slobodan Djurdjevic, Hieu Nguyen, Robert Slater

SMU Data Science Review

The problem of forecasting market volatility is a difficult task for most fund managers. Volatility forecasts are used for risk management, alpha (risk) trading, and the reduction of trading friction. Improving the forecasts of future market volatility assists fund managers in adding or reducing risk in their portfolios as well as in increasing hedges to protect their portfolios in anticipation of a market sell-off event. Our analysis compares three existing financial models that forecast future market volatility using the Chicago Board Options Exchange Volatility Index (VIX) to six machine/deep learning supervised regression methods. This analysis determines which models provide best …

Intelligent Technology Of Command And Control System In The Rts Perspective, Wenfeng Wu, Zhang Yu, Rong Ming

Journal of System Simulation

Abstract: Real-Time Strategy (RTS) games have important reference value for studying the intelligent technology of command and control systems. The similarities between RTS games and the strategic battle level command and control systems are described according to the decision process. The challenges brought by the problems of planning, learning, uncertainty and space-time reasoning in the intelligent technology of RTS games are analyzed. The key technologies and latest research progress of action sequence planning, plan recognition, state assessment, multi-agent collaboration and multi-scale AI are studied. The trend of intelligent technology development of strategic and operational level command and control systems is …

Deep Learning: Edge-Cloud Data Analytics For Iot, Katarina Grolinger, Ananda M. Ghosh

Electrical and Computer Engineering Publications

Sensors, wearables, mobile and other Internet of Thing (IoT) devices are becoming increasingly integrated in all aspects of our lives. They are capable of collecting massive quantities of data that are typically transmitted to the cloud for processing. However, this results in increased network traffic and latencies. Edge computing has a potential to remedy these challenges by moving computation physically closer to the network edge where data are generated. However, edge computing does not have sufficient resources for complex data analytics tasks. Consequently, this paper investigates merging cloud and edge computing for IoT data analytics and presents a deep learning-based …

Emotion Forecasting In Dyadic Conversation : Characterizing And Predicting Future Emotion With Audio-Visual Information Using Deep Learning, Sadat Shahriar

Legacy Theses & Dissertations (2009 - 2024)

Emotion forecasting is the task of predicting the future emotion of a speaker, i.e., the emotion label of the future speaking turn–based on the speaker’s past and current audio-visual cues. Emotion forecasting systems require new problem formulations that differ from traditional emotion recognition systems. In this thesis, we first explore two types of forecasting windows(i.e., analysis windows for which the speaker’s emotion is being forecasted): utterance forecasting and time forecasting. Utterance forecasting is based on speaking turns and forecasts what the speaker’s emotion will be after one, two, or three speaking turns. Time forecasting forecasts what the speaker’s emotion will …

Multi-Spectral Visual Crop Assessment Under Limited Data Constraints, Patricia O'Byrne, Patrick Jackman, Damon Berry, Hector-Hugo Franco-Penya, Michael French, Robert J. Ross

Conference papers

In an era of climate change and global population growth, deep learning based multi-spectral imaging has the potential to significantly assist in production management across a wide range of agricultural and food production domains. A key challenge however in applying state-of-the-art methods is that they, unlike classical hand crafted methods, are usually thought of as being only useful when significant amounts of data are available. In this paper we investigate this hypothesis by examining the performance of state-of-the-art deep learning methods when applied to a restricted data set that is not easily bootstrapped through pre-trained image processing networks. We demonstrate …