Deep learning (Machine learning)

The rise of Advanced Driver-Assistance Systems (ADAS) and Autonomous Vehicles (AVs) emphasizes the urgent need to combat distracted driving. This study introduces a fresh approach for improved detection of distracted drivers, combining a pre-trained Convolutional Neural Network (CNN) with a Bidirectional Long Short- Term Memory (BiLSTM) network. Our analysis utilizes both spatial and temporal features to examine a broad array of driver distractions. We demonstrate the advantage of this CNN-BiLSTM framework over conventional methods, achieving significant precision (up to 98.97%) on the combined ’Union Dataset,’ merging the Kaggle State Farm Dataset and AUC Distracted Driver Dataset (AUC-DDD). This research enhances safety in autonomous vehicles by providing a solid and flexible solution for everyday use. Our results mark considerable progress in accurately identifying driver distractions, pushing the boundaries of safety technology in AVs.

With the recent large-scale adoption of Large Language Models in multidisciplinary research and commercial space, the need for large amounts of labeled data has become more crucial than ever to evaluate potential use cases for opportunities in applied intelligence. Most domain specific fields require a substantial shift that involves extremely large amounts of heterogeneous data to have meaningful impact on the pre-computed weights of most large language models. We explore extending the capabilities a state-of-the-art unsupervised pre-training method; Transformers and Sequential Denoising Auto-Encoder (TSDAE). In this study we show various opportunities for using OCR2Seq a multi-modal generative augmentation strategy to further enhance and measure the quality of noise samples used when using TSDAE as a pretraining task. This study is a first of its kind work that leverages converting both generalized and sparse domains of relational data into multi-modal sources. Our primary objective is measuring the quality of augmentation in relation to the current implementation of the sentence transformers library. Further work includes the effect on ranking, language understanding, and corrective quality.

Online advertising [100], as a multi-billion dollar business, provides a common marketing experience when people access online services using electronic devices, such as desktop computers, tablets, smartphones, and so on. Using the Internet as a means of advertising, different stakeholders take actions in the background to provide and deliver advertisements to users through numerous platforms, such as search engines, news sites, and social networks, where dedicated spots of areas are used to display advertisements (ads) along with search results, posts, or page content.
Online advertising is mainly based on dynamically selecting ads through a real-time bidding (or auction) mechanism. Predicting user responses like clicking ads in e-commerce platforms and internet-based advertising systems, as the first measurable user response, is an essential step for many digital advertising and recommendation systems to capture the user’s propensity to follow up actions, such as purchasing a product or subscribing to a service. To maximize revenue and user satisfaction, online advertising platforms must predict the expected user behavior of each displayed advertisement and maximize the user’s expectations of clicking [28]. Based on this observed feedback, these systems are tailored to user preferences to decide the order in that ads or any promoted content should be served to them. This objective provides an incentive to develop new research by using ideas derived from different domains like machine learning and data mining combined with models for information retrieval and mathematical optimization. They introduce different machine learning and data mining methods that employ deep learning-based predictive models to learn the representation of input features with the aim of user response prediction. Feature representation learning is known as a fundamental task on how to input information is going to be represented in machine learning models. A good feature representation learning method that seeks to learn low-dimensional embedding vectors is a key factor for the success of many downstream analytics tasks, such as click-through prediction and conversion prediction in recommendation systems and online advertising platforms.

Skin cancer is a prevalent cancer that significantly contributes to global mortality rates. Early detection is crucial for a high survival rate. Dermatologists primarily rely on visual inspection to diagnose skin cancers, but this method is inaccurate. Deep learning algorithms can enhance the diagnostic accuracy of skin cancers. However, these algorithms require substantial labeled data for effective training. Acquiring annotated data for skin cancer classification is time-consuming, expensive, and necessitates expert annotation. Moreover, skin cancer datasets often suffer from imbalanced data distribution.
Generative Adversarial Networks (GANs) can be used to overcome the challenges of data scarcity and lack of labels by automatically generating skin cancer images. However, training and testing data from different distributions can introduce domain shift and bias, impacting the model’s performance. This dissertation addresses this issue by developing deep learning-based domain adaptation models.
Additionally, this research emphasizes deploying deep learning models on hardware to enable real-time skin cancer detection, facilitating accurate diagnoses by dermatologists. Deploying conventional deep learning algorithms on hardware is not preferred due to the problem of high resource consumption. Therefore, this dissertation presents spiking neural network-based (SNN) models designed specifically for hardware implementation. SNNs are preferred for their power-efficient behavior and suitability for hardware deployment.

The major objective of this dissertation was to create a framework which is used for medical image diagnosis. In this diagnosis, we brought classification and diagnosing of diseases through an Artificial Intelligence based framework, including COVID, Pneumonia, and Melanoma cancer through medical images. The algorithm ran on multiple datasets. A model was developed which detected the medical images through changing hyper-parameters.
The aim of this work was to apply the new transfer learning framework DenseNet-201 for the diagnosis of the diseases and compare the results with the other deep learning models. The novelty in the proposed work was modifying the Dense Net 201 Algorithm, changing hyper parameters (source weights, Batch Size, Epochs, Architecture (number of neurons in hidden layer), learning rate and optimizer) to quantify the results. The novelty also included the training of the model by quantifying weights and in order to get more accuracy. During the data selection process, the data were cleaned, removing all the outliers. Data augmentation was used for the novel architecture to overcome overfitting and hence not producing false absurd results the computational performance was also observed. The proposed model results were also compared with the existing deep learning models and the algorithm was also tested on multiple datasets.

Recent successes of Deep Learning-powered AI are largely due to the trio of: algorithms, GPU computing, and big data. Data could take the shape of hospital records, satellite images, or the text in this paragraph. Deep Learning algorithms typically need massive collections of data before they can make reliable predictions. This limitation inspired investigation into a class of techniques referred to as Data Augmentation. Data Augmentation was originally developed as a set of label-preserving transformations used in order to simulate large datasets from small ones. For example, imagine developing a classifier that categorizes images as either a “cat” or a “dog”. After initial collection and labeling, there may only be 500 of these images, which are not enough data points to train a Deep Learning model. By transforming these images with Data Augmentations such as rotations and brightness modifications, more labeled images are available for model training and classification! In addition to applications for learning from limited labeled data, Data Augmentation can also be used for generalization testing. For example, we can augment the test set to set the visual style of images to “winter” and see how that impacts the performance of a stop sign detector.
The dissertation begins with an overview of Deep Learning methods such as neural network architectures, gradient descent optimization, and generalization testing. Following an initial description of this technology, the dissertation explains overfitting. Overfitting is the crux of Deep Learning methods in which improvements to the training set do not lead to improvements on the testing set. To the rescue are Data Augmentation techniques, of which the Dissertation presents an overview of the augmentations used for both image and text data, as well as the promising potential of generative data augmentation with models such as ChatGPT. The dissertation then describes three major experimental works revolving around CIFAR-10 image classification, language modeling a novel dataset of Keras information, and patient survival classification from COVID-19 Electronic Health Records. The dissertation concludes with a reflection on the evolution of limitations of Deep Learning and directions for future work.

Cyber attack is a strong threat to the digital world. So, it’s very essential to keep the network safe. Network Intrusion Detection system is the system to address this problem. Network Intrusion Detection system functions like a firewall, and monitors incoming and outgoing traffic like ingress and egress filtering fire wall. Network Intrusion Detection System does anomaly and hybrid detection for detecting known and unknown attacks. My thesis discusses about the several network cyber attacks we face nowadays and I created several Deep learning models to detect accurately, I used NSL-KDD dataset which is a popular dataset, that contains several network attacks. After experimenting with different deep learning models I found some disparities in the training accuracy and validation accuracy, which is a clear indication of overfitting. To reduce the overfitting I introduced regularization and dropout in the models and experimented with different hyperparameters.

Wall-bounded turbulent flows are pervasive in numerous physics and engineering applications. Such flows tend to have a strong impact on the design of ships, airplanes and rockets, industrial chemical mixing, wind and hydrokinetic energy, utility infrastructure and innumerable other fields. Understanding and controlling wall bounded turbulence has been a long-pursued endeavor yielding plentiful scientific and engineering discoveries, but there is much that remains unexplained from a fundamental viewpoint. One unexplained phenomenon is the formation and impact of coherent structures like the ejections of slow near-wall fluid into faster moving ow which have been shown to correlate with increases in friction drag. This thesis focuses on recognizing and regulating organized structures within wall-bounded turbulent flows using a variety of machine learning techniques to overcome the nonlinear nature of this phenomenon.
Deep Learning has provided new avenues of analyzing large amounts of data by applying techniques modeled after biological neurons. These techniques allow for the discovery of nonlinear relationships in massive, complex systems like the data found frequently in fluid dynamics simulation. Using a neural network architecture called Convolutional Neural Networks that specializes in uncovering spatial relationships, a network was trained to estimate the relative intensity of ejection structures within turbulent flow simulation without any a priori knowledge of the underlying flow dynamics. To explore the underlying physics that the trained network might reveal, an interpretation technique called Gradient-based Class Activation Mapping was modified to identify salient regions in the flow field which most influenced the trained network to make an accurate estimation of these organized structures. Using various statistical techniques, these salient regions were found to have a high correlation to ejection structures, and to high positive kinetic energy production, low negative production, and low energy dissipation regions within the flow. Additionally, these techniques present a general framework for identifying nonlinear causal structures in general three-dimensional data in any scientific domain where the underlying physics may be unknown.

The field of computer vision has grown by leaps and bounds in the past decade. The rapid advances can be largely attributed to advances made in the field of Artificial Neural Networks and more specifically can be attributed to the rapid advancement of Convolutional Neural Networks (CNN) and Deep Learning. One area that is of great interest to the research community at large is the ability to detect the quality of images in the sense of technical parameters such as blurriness, encoding artifacts, saturation, and lighting, as well as for its’ aesthetic appeal. The purpose of such a mechanism could be detecting and discarding noisy, blurry, dark, or over exposed images, as well as detecting images that would be considered beautiful by a majority of viewers. In this dissertation, the detection of various quality and aesthetic aspects of an image using CNNs is explored. This research produced two datasets that are manually labeled for quality issues such as blur, poor lighting, and digital noise, and for their aesthetic qualities, and Convolutional Neural Networks were designed and trained using these datasets. Lastly, two case studies were performed to show the real-world impact of this research to traffic sign detection and medical image diagnosis.

Time-series data in biomedical applications are gaining an increased interest to detect and predict underlying diseases and estimate their severity, such as Parkinson’s disease (PD) and cardiovascular diseases. This interest is driven by advances in wearable sensors and deep learning models to a large extent. In the literature, less attention has been paid to regression models for continuous outcomes in these applications, especially when dealing with limited data. Training deep learning models on raw limited data results in overfitted models, which is the main technical challenge we address in this dissertation. An example of limited and\or imbalanced time-series data is PD’s motion signals that are needed for the continuous severity estimation of Parkinson’s disease (PD). The significance of this continuous estimation is providing a tool for longitudinal monitoring of daily motor and non-motor fluctuations and managing PD medications.
The dissertation objective is to train generalizable deep learning models for biomedical regression problems when dealing with limited training time-series data. The goal is designing, developing, and validating an automatic assessment system based on wearable sensors that can measure the severity of PD complications in the home-living environment while patients with PD perform their activities of daily living (ADL). We first propose using a combination of domain-specific feature engineering, transfer learning, and an ensemble of multiple modalities. Second, we utilize generative adversarial networks (GAN) and propose a new formulation of conditional GAN (cGAN) as a generative model for regression to handle an imbalanced training dataset. Next, we propose a dual-channel auxiliary regressor GAN (AR-GAN) trained using Wasserstein-MSE-correlation loss. The proposed AR-GAN is used as a data augmentation method in regression problems.