Physical Sciences and Mathematics Commons^™

End-To-End Deep Reinforcement Learning For Multi-Agent Collaborative Exploration, Zichen Chen, Budhitama Subagdja, Ah-Hwee Tan

Research Collection School Of Computing and Information Systems

Exploring an unknown environment by multiple autonomous robots is a major challenge in robotics domains. As multiple robots are assigned to explore different locations, they may interfere each other making the overall tasks less efficient. In this paper, we present a new model called CNN-based Multi-agent Proximal Policy Optimization (CMAPPO) to multi-agent exploration wherein the agents learn the effective strategy to allocate and explore the environment using a new deep reinforcement learning architecture. The model combines convolutional neural network to process multi-channel visual inputs, curriculum-based learning, and PPO algorithm for motivation based reinforcement learning. Evaluations show that the proposed method …

Fusion Of Multimodal Embeddings For Ad-Hoc Video Search, Danny Francis, Phuong Anh Nguyen, Benoit Huet, Chong-Wah Ngo

Research Collection School Of Computing and Information Systems

The challenge of Ad-Hoc Video Search (AVS) originates from free-form (i.e., no pre-defined vocabulary) and freestyle (i.e., natural language) query description. Bridging the semantic gap between AVS queries and videos becomes highly difficult as evidenced from the low retrieval accuracy of AVS benchmarking in TRECVID. In this paper, we study a new method to fuse multimodal embeddings which have been derived based on completely disjoint datasets. This method is tested on two datasets for two distinct tasks: on MSR-VTT for unique video retrieval and on V3C1 for multiple videos retrieval.

On Video Analysis Of Omnidirectional Bee Traffic: Counting Bee Motions With Motion Detection And Image Classification, Vladmir Kulyukin, Sarbajit Mukherjee

Computer Science Faculty and Staff Publications

Omnidirectional bee traffic is the number of bees moving in arbitrary directions in close proximity to the landing pad of a given hive over a given period of time. Video bee traffic analysis has the potential to automate the assessment of omnidirectional bee traffic levels, which, in turn, may lead to a complete or partial automation of honeybee colony health assessment. In this investigation, we proposed, implemented, and partially evaluated a two-tier method for counting bee motions to estimate levels of omnidirectional bee traffic in bee traffic videos. Our method couples motion detection with image classification so that motion detection …

Semi-Supervised Regression With Generative Adversarial Networks Using Minimal Labeled Data, Greg Olmschenk

Dissertations, Theses, and Capstone Projects

This work studies the generalization of semi-supervised generative adversarial networks (GANs) to regression tasks. A novel feature layer contrasting optimization function, in conjunction with a feature matching optimization, allows the adversarial network to learn from unannotated data and thereby reduce the number of labels required to train a predictive network. An analysis of simulated training conditions is performed to explore the capabilities and limitations of the method. In concert with the semi-supervised regression GANs, an improved label topology and upsampling technique for multi-target regression tasks are shown to reduce data requirements. Improvements are demonstrated on a wide variety of vision …

Rotation Invariant Convolutions For 3d Point Clouds Deep Learning, Zhiyuan Zhang, Binh-Son Hua, David W. Rosen, Sai-Kit Yeung

Research Collection School Of Computing and Information Systems

Recent progresses in 3D deep learning has shown that it is possible to design special convolution operators to consume point cloud data. However, a typical drawback is that rotation invariance is often not guaranteed, resulting in networks that generalizes poorly to arbitrary rotations. In this paper, we introduce a novel convolution operator for point clouds that achieves rotation invariance. Our core idea is to use low-level rotation invariant geometric features such as distances and angles to design a convolution operator for point cloud learning. The well-known point ordering problem is also addressed by a binning approach seamlessly built into the …

Exploiting Approximation, Caching And Specialization To Accelerate Vision Sensing Applications, Nguyen Loc Huynh

Dissertations and Theses Collection (Open Access)

Over the past few years, deep learning has emerged as state-of-the-art solutions for many challenging computer vision tasks such as face recognition, object detection, etc. Despite of its outstanding performance, deep neural networks (DNNs) are computational intensive, which prevent them to be widely adopted on billions of mobile and embedded devices with scarce resources. To address that limitation, we
focus on building systems and optimization algorithms to accelerate those models, making them more computational-efficient.
First, this thesis explores the computational capabilities of different existing processors (or co-processors) on modern mobile devices. It recognizes that by leveraging the mobile Graphics Processing …

Applied Deep Learning In Intelligent Transportation Systems And Embedding Exploration, Xiaoyuan Liang

Dissertations

Deep learning techniques have achieved tremendous success in many real applications in recent years and show their great potential in many areas including transportation. Even though transportation becomes increasingly indispensable in people’s daily life, its related problems, such as traffic congestion and energy waste, have not been completely solved, yet some problems have become even more critical. This dissertation focuses on solving the following fundamental problems: (1) passenger demand prediction, (2) transportation mode detection, (3) traffic light control, in the transportation field using deep learning. The dissertation also extends the application of deep learning to an embedding system for visualization …

Learning For Clinical Outcome Prediction From Big Medical Data, Jiawen Yao

Computer Science and Engineering Dissertations - Archive

With the advance of recent technological innovations, nowadays scientists can easily capture and store tremendous amounts of different types of medical data such as Computed Tomography (CT), Magnetic Resonance Imaging (MRI), big pathological images and high dimensional cell profiling data. Developing deep learning and machine learning to analyze such large-scale medical data sets for patient health care is an interesting but challenging problem. Inspired by the trend, in this dissertation, we focus on solving real-world problems, like survival analysis on image-omics data and reducing uncertainty from undersampled MRI. Survival analysis is a crucial tool in the clinical study of cancer …

Convex And Non-Convex Optimization Methods For Machine Learning, Fariba Zohrizadeh

Computer Science and Engineering Dissertations - Archive

This dissertation is concerned with modeling fundamental and challenging machine learning tasks as convex/non-convex optimization problems and designing a mechanism that could solve them in a cost and time-effective manner. Extensive theoretical and practical studies are carried out to give deeper insights into the robustness and effectiveness of the formulated problems. In what follows, we investigate some well-known tasks that frequently arise in machine learning applications. Image Segmentation: Image segmentation is a fundamental and challenging task in computer vision with diverse applications in various areas. One of the major challenges in image segmentation is to determine the optimal number of …

Deepstellar: Model-Based Quantitative Analysis Of Stateful Deep Learning Systems, Xiaoning Du, Xiaofei Xie, Yi Li, Lei Ma, Yang Liu, Jianjun Zhao

Research Collection School Of Computing and Information Systems

Deep Learning (DL) has achieved tremendous success in many cutting-edge applications. However, the state-of-the-art DL systems still suffer from quality issues. While some recent progress has been made on the analysis of feed-forward DL systems, little study has been done on the Recurrent Neural Network (RNN)-based stateful DL systems, which are widely used in audio, natural languages and video processing, etc. In this paper, we initiate the very first step towards the quantitative analysis of RNN-based DL systems. We model RNN as an abstract state transition system to characterize its internal behaviors. Based on the abstract model, we design two …

Analysis Of The Cnn Algorithm In Target Recognition By Using The Mstar Database, Ligang Zou

Theses and Dissertations

With the rapid development of artificial intelligence technology and the emergence of a large number of innovative theories, the concept of deep learning is widely used in object detection, speech recognition, language translation and other fields. One of the important practices is target recognition in SAR images. Although it shows certain effectiveness in some researches, when using deep learning algorithm, there are still many problems that have not yet been solved. For example, people do not have a good understanding of how convolution works and the impact of convolution on the algorithm, although convolution works well in the CNN algorithm. …

Deep Learning For Crack-Like Object Detection, Kaige Zhang

All Graduate Theses and Dissertations, Spring 1920 to Summer 2023

Cracks are common defects on surfaces of man-made structures such as pavements, bridges, walls of nuclear power plants, ceilings of tunnels, etc. Timely discovering and repairing of the cracks are of great significance and importance for keeping healthy infrastructures and preventing further damages. Traditionally, the cracking inspection was conducted manually which was labor-intensive, time-consuming and costly. For example, statistics from the Central Intelligence Agency show that the world’s road network length has reached 64,285,009 km, of which the United States has 6,586,610 km. It is a huge cost to maintain and upgrade such an immense road network. Thus, fully automatic …

Identifying Depression In The National Health And Nutrition Examination Survey Data Using A Deep Learning Algorithm, Jihoon Oh, Kyongsik Yun, Uri Maoz, Tae-Suk Kim, Jeong-Ho Chae

Psychology Faculty Articles and Research

Background

As depression is the leading cause of disability worldwide, large-scale surveys have been conducted to establish the occurrence and risk factors of depression. However, accurately estimating epidemiological factors leading up to depression has remained challenging. Deep-learning algorithms can be applied to assess the factors leading up to prevalence and clinical manifestations of depression.

Methods

Customized deep-neural-network and machine-learning classifiers were assessed using survey data from 19,725 participants from the NHANES database (from 1999 through 2014) and 4949 from the South Korea NHANES (K-NHANES) database in 2014.

Results

A deep-learning algorithm showed area under the receiver operating characteristic curve (AUCs) …

Incremental Data Annotation With Mask R-Cnn, Michael Miller

Theses and Dissertations

With the advent of object detection and instance segmentation algorithms, the complexity of annotating training data has compounded. In traditional image classification tasks, training images must only be given a label. This type of training data alone is not enough to perform detection or instance segmentation as we need additional object level information like bounding boxes or segmentation masks appended to the annotation. There are many tools that help a user process images to produce data annotations that contain bounding box and segmentation mask information. However, most of these are semi-supervised and require a user to be present for each …

Integration Of Random Forest Classifiers And Deep Convolutional Neural Networks For Classification And Biomolecular Modeling Of Cancer Driver Mutations, Steve Agajanian, Odeyemi Oluyemi, Gennady M. Verkhivker

Mathematics, Physics, and Computer Science Faculty Articles and Research

Development of machine learning solutions for prediction of functional and clinical significance of cancer driver genes and mutations are paramount in modern biomedical research and have gained a significant momentum in a recent decade. In this work, we integrate different machine learning approaches, including tree based methods, random forest and gradient boosted tree (GBT) classifiers along with deep convolutional neural networks (CNN) for prediction of cancer driver mutations in the genomic datasets. The feasibility of CNN in using raw nucleotide sequences for classification of cancer driver mutations was initially explored by employing label encoding, one hot encoding, and embedding to …

Corrn: Cooperative Reflection Removal Network, Renjie Wen, Boxin Shi, Haoliang Li, Ling-Yu Duan, Ah-Hwee Tan, Alex C. Kot

Research Collection School Of Computing and Information Systems

Removing the undesired reflections from images taken through the glass is of broad application to various computer vision tasks. Non-learning based methods utilize different handcrafted priors such as the separable sparse gradients caused by different levels of blurs, which often fail due to their limited description capability to the properties of real-world reflections. In this paper, we propose a network with the feature-sharing strategy to tackle this problem in a cooperative and unified framework, by integrating image context information and the multi-scale gradient information. To remove the strong reflections existed in some local regions, we propose a statistic loss by …

Model-Based Deep Autoencoders For Characterizing Discrete Data With Application To Genomic Data Analysis, Tian Tian

Dissertations

Deep learning techniques have achieved tremendous successes in a wide range of real applications in recent years. For dimension reduction, deep neural networks (DNNs) provide a natural choice to parameterize a non-linear transforming function that maps the original high dimensional data to a lower dimensional latent space. Autoencoder is a kind of DNNs used to learn efficient feature representation in an unsupervised manner. Deep autoencoder has been widely explored and applied to analysis of continuous data, while it is understudied for characterizing discrete data. This dissertation focuses on developing model-based deep autoencoders for modeling discrete data. A motivating example of …

Statistical Machine Learning Methods For Mining Spatial And Temporal Data, Fei Tan

Dissertations

Spatial and temporal dependencies are ubiquitous properties of data in numerous domains. The popularity of spatial and temporal data mining has thus grown with the increasing prevalence of massive data. The presence of spatial and temporal attributes not only provides complementary useful perspectives, but also poses new challenges to the representation and integration into the learning procedure. In this dissertation, the involved spatial and temporal dependencies are explored with three genres: sample-wise, feature-wise, and target-wise. A family of novel methodologies is developed accordingly for the dependency representation in respective scenarios.

First, dependencies among discrete, continuous and repeated observations are studied …

Deep Morphological Neural Networks, Yucong Shen

Theses

Mathematical morphology is a theory and technique applied to collect features like geometric and topological structures in digital images. Determining suitable morphological operations and structuring elements for a give purpose is a cumbersome and time-consuming task. In this paper, morphological neural networks are proposed to address this problem. Serving as a non-linear feature extracting layers in deep learning frameworks, the efficiency of the proposed morphological layer is confirmed analytically and empirically. With a known target, a single-filter morphological layer learns the structuring element correctly, and an adaptive layer can automatically select appropriate morphological operations. For high level applications, the proposed …

Music Mood Classification Using Convolutional Neural Networks, Revanth Akella

Master's Projects

Grouping music into moods is useful as music is migrating from to online streaming services as it can help in recommendations. To establish the connection between music and mood we develop an end-to-end, open source approach for mood classification using lyrics. We develop a pipeline for tag extraction, lyric extraction, and establishing classification models for classifying music into moods. We investigate techniques to classify music into moods using lyrics and audio features. Using various natural language processing methods with machine learning and deep learning we perform a comparative study across different classification and mood models. The results infer that features …

Using Computer Vision To Quantify Coral Reef Biodiversity, Niket Bhodia

Master's Projects

The preservation of the world’s oceans is crucial to human survival on this planet, yet we know too little to begin to understand anthropogenic impacts on marine life. This is especially true for coral reefs, which are the most diverse marine habitat per unit area (if not overall) as well as the most sensitive. To address this gap in knowledge, simple field devices called autonomous reef monitoring structures (ARMS) have been developed, which provide standardized samples of life from these complex ecosystems. ARMS have now become successful to the point that the amount of data collected through them has outstripped …

Intelligent Log Analysis For Anomaly Detection, Steven Yen

Master's Projects

Computer logs are a rich source of information that can be analyzed to detect various issues. The large volumes of logs limit the effectiveness of manual approaches to log analysis. The earliest automated log analysis tools take a rule-based approach, which can only detect known issues with existing rules. On the other hand, anomaly detection approaches can detect new or unknown issues. This is achieved by looking for unusual behavior different from the norm, often utilizing machine learning (ML) or deep learning (DL) models. In this project, we evaluated various ML and DL techniques used for log anomaly detection. We …

Robust Lightweight Object Detection, Siddharth Kumar

Master's Projects

Object detection is a very challenging problem in computer vision and has been a prominent subject of research for nearly three decades. There has been a promising in- crease in the accuracy and performance of object detectors ever since deep convolutional networks (CNN) were introduced. CNNs can be trained on large datasets made of high resolution images without flattening them, thereby using the spatial information. Their superior learning ability also makes them ideal for image classification and object de- tection tasks. Unfortunately, this power comes at the big cost of compute and memory. For instance, the Faster R-CNN detector required …

Machine Learning Versus Deep Learning For Malware Detection, Parth Jain

Master's Projects

It is often claimed that the primary advantage of deep learning is that such models can continue to learn as more data is available, provided that sufficient computing power is available for training. In contrast, for other forms of machine learning it is claimed that models ‘‘saturate,’’ in the sense that no additional learning can occur beyond some point, regardless of the amount of data or computing power available. In this research, we compare the accuracy of deep learning to other forms of machine learning for malware detection, as a function of the training dataset size. We experiment with a …

Chatbots With Personality Using Deep Learning, Susmit Gaikwad

Master's Projects

Natural Language Processing (NLP) requires the computational modelling of the complex relationships of the syntax and semantics of a language. While traditional machine learning methods are used to solve NLP problems, they cannot imitate the human ability for language comprehension. With the growth in deep learning, these complexities within NLP are easier to model, and be used to build many computer applications. A particular example of this is a chatbot, where a human user has a conversation with a computer program, that generates responses based on the user’s input. In this project, we study the methods used in building chatbots, …

Optimizing E-Commerce Product Classification Using Transfer Learning, Rashmeet Kaur Khanuja

Master's Projects

The global e-commerce market is snowballing at a rate of 23% per year. In 2017, retail e-commerce users were 1.66 billion and sales worldwide amounted to 2.3 trillion US dollars, and e-retail revenues are projected to grow to 4.88 trillion USD in 2021. With the immense popularity that e-commerce has gained over past few years comes the responsibility to deliver relevant results to provide rich user experience. In order to do this, it is essential that the products on the ecommerce website be organized correctly into their respective categories. Misclassification of products leads to irrelevant results for users which not …

Image Retrieval Using Image Captioning, Nivetha Vijayaraju

Master's Projects

The rapid growth in the availability of the Internet and smartphones have resulted in the increase in usage of social media in recent years. This increased usage has thereby resulted in the exponential growth of digital images which are available. Therefore, image retrieval systems play a major role in fetching images relevant to the query provided by the users. These systems should also be able to handle the massive growth of data and take advantage of the emerging technologies, like deep learning and image captioning. This report aims at understanding the purpose of image retrieval and various research held in …

Stock Market Prediction Using Ensemble Of Graph Theory, Machine Learning And Deep Learning Models, Pratik Patil

Master's Projects

Efficient Market Hypothesis (EMH) is the cornerstone of the modern financial theory and it states that it is impossible to predict the price of any stock using any trend, fundamental or technical analysis. Stock trading is one of the most important activities in the world of finance. Stock price prediction has been an age-old problem and many researchers from academia and business have tried to solve it using many techniques ranging from basic statistics to machine learning using relevant information such as news sentiment and historical prices. Even though some studies claim to get prediction accuracy higher than a random …

Deep Learning For Image Spam Detection, Tazmina Sharmin

Master's Projects

Spam can be defined as unsolicited bulk email. In an effort to evade text-based spam filters, spammers can embed their spam text in an image, which is referred to as image spam. In this research, we consider the problem of image spam detection, based on image analysis. We apply various machine learning and deep learning techniques to real-world image spam datasets, and to a challenge image spam-like dataset. We obtain results comparable to previous work for the real-world datasets, while our deep learning approach yields the best results to date for the challenge dataset.

Selecting Maximally-Predictive Deep Features To Explain What Drives Fixations In Free-Viewing, Matthias Kümmerer, Thomas S.A. Wallis, Matthias Bethge

MODVIS Workshop

No abstract provided.