Explaining Deep Q-Learning Experience Replay With Shapley Additive Explanations,
2023
Technological University Dublin, Ireland
Explaining Deep Q-Learning Experience Replay With Shapley Additive Explanations, Robert S. Sullivan
Dissertations
Reinforcement Learning (RL) has shown promise in optimizing complex control and decision-making processes but Deep Reinforcement Learning (DRL) lacks interpretability, limiting its adoption in regulated sectors like manufacturing, finance, and healthcare. Difficulties arise from DRL’s opaque decision-making, hindering efficiency and resource use, this issue is amplified with every advancement. While many seek to move from Experience Replay to A3C, the latter demands more resources. Despite efforts to improve Experience Replay selection strategies, there is a tendency to keep capacity high. This dissertation investigates training a Deep Convolutional Q-learning agent across 20 Atari games, in solving a control task, physics task, …
The Use Of Data Balancing Algorithms To Correct For The Under-Representation Of Female Patients In A Cardiovascular Dataset,
2023
Technological University Dublin, Ireland
The Use Of Data Balancing Algorithms To Correct For The Under-Representation Of Female Patients In A Cardiovascular Dataset, Sian Miller
Dissertations
Given that women are under-represented in medical datasets, and that machine learning classification algorithms are known to exhibit bias towards the majority class, the growing application of machine learning in the medical field risks resulting in worse medical outcomes for female patients. The Heart Failure Prediction (HFP) dataset is a historical dataset used for the training of models for the prediction of heart disease. This dataset contains significantly fewer female patients than male patients, and as such it is expected that models trained using this data will inherit a gender bias to favour male patients. This dissertation explores the use …
Probability Expressions In Ai Decision Support: Impacts On Human+Ai Team Performance,
2023
Technological University Dublin, Ireland
Probability Expressions In Ai Decision Support: Impacts On Human+Ai Team Performance, Elias Spinn
Dissertations
AI decision support systems aim to assist people in highly complex and consequential domains to make efficient, effective, and high-quality decisions. AI alone cannot be guaranteed to be correct in these complex decision tasks, and a human is often needed to ensure decision accuracy. The ambition is for these human+ AI teams to perform better together than either would individually. To realise this, decision makers must trust their AI partners appropriately, knowing when to rely on their recommendations and when to be sceptical. However, research has shown that decision makers often either mistrust and underutilise these systems, or trust them …
Modelling And Mitigating Interlocutor Confusion In Situated Human-Avatar And Human-Robot Interaction,
2023
Technological University Dublin
Modelling And Mitigating Interlocutor Confusion In Situated Human-Avatar And Human-Robot Interaction, Na Li
Dissertations
Human-Robot Interaction (HRI) is an important but challenging field focused on improving the interaction between humans and robots, to make the interaction more intelligent and effective. However, building a natural conversational HRI is an interdisciplinary challenge for scholars, engineers, and designers. Achieving successful conversational interaction with a social robot necessitates not only observing a user’s active participation in the interaction but also being aware of their emotional and attitudinal states as the interaction progresses. On the topic of attitudinal states, one field that has received little attention to date is monitoring the user for possible confusion states. Confusion is a …
Demo Alleviate: Demonstrating Artificial Intelligence Enabled Virtual Assistance For Telehealth: The Mental Health Case,
2023
University of South Carolina - Columbia
Demo Alleviate: Demonstrating Artificial Intelligence Enabled Virtual Assistance For Telehealth: The Mental Health Case, Kaushik Roy, Vedant Khandelwal, Raxit Goswami, Nathan Dolbir, Jinendra Malekar, Amit Sheth
Publications
After the pandemic, artificial intelligence (AI) powered support for mental health care has become increasingly important. The breadth and complexity of significant challenges required to provide adequate care involve: (a) Personalized patient understanding, (b) Safety-constrained and medically validated chatbot patient interactions, and (c) Support for continued feedback-based refinements in design using chatbot-patient interactions. We propose Alleviate, a chatbot designed to assist patients suffering from mental health challenges with personalized care and assist clinicians with understanding their patients better. Alleviate draws from an array of publicly available clinically valid mental-health texts and databases, allowing Alleviate to make medically sound and informed …
Impact Of Character N-Grams Attention Scores For English And Russian News Articles Authorship Attribution,
2023
Technological University Dublin, Ireland
Impact Of Character N-Grams Attention Scores For English And Russian News Articles Authorship Attribution, Liliya Mukhmutova, Robert J. Ross, Giancarlo Salton
Conference papers
Language embeddings are often used as black-box word-level tools that provide powerful language analysis across many tasks, but yet for many tasks such as Authorship Attribution access to feature level information on character n-grams can provide insights to help with model refinement and development. In this paper we investigate and evaluate the importance of character n-grams within an embeddings context in authorship attribution through the use of attention scores. We perform this investigation both for English (Reuters_50_50) and Russian (Taiga) news authorship datasets. Our analysis show that character n-grams attention score is higher for n-grams that are considered to be …
Subnetwork Ensembling And Data Augmentation: Effects On Calibration,
2023
Technological University Dublin, Ireland
Subnetwork Ensembling And Data Augmentation: Effects On Calibration, A. Çağrı Demir, Simon Caton, Pierpaolo Dondio
Articles
Deep Learning models based on convolutional neural networks are known to be uncalibrated, that is, they are either overconfident or underconfident in their predictions. Safety-critical applications of neural networks, however, require models to be well-calibrated, and there are various methods in the literature to increase model performance and calibration. Subnetwork ensembling is based on the over-parametrization of modern neural networks by fitting several subnetworks into a single network to take advantage of ensembling them without additional computational costs. Data augmentation methods have also been shown to enhance model performance in terms of accuracy and calibration. However, ensembling and data augmentation …
Gated Deep Reinforcement Learning With Red Deer Optimization For Medical Image Classification,
2023
Vellore Institute of Technology, Chennai, India
Gated Deep Reinforcement Learning With Red Deer Optimization For Medical Image Classification, Narayanan Ganesh, Sambandan Jayalakshmi, Rama Chandran Narayanan, Miroslav Mahdal, Hossam Zawbaa, Ali Wagdy Mohamed
Articles
The brain is one of the most important and complex organs in the body, consisting of billions of individual cells. Uncontrolled growth and expansion of aberrant cell populations within or around the brain are the main causes of brain tumors. These cells have the potential to harm healthy cells and impair brain function [1]. Tumors can be detected using medical imaging techniques, which are considered the most popular and accurate way to classify different types of cancer, and this procedure is even more crucial as it is noninvasive [2]. Magnetic resonance imaging (MRI) is one such medical imaging technique that …
Ontology-Based Case Study Management Towards Bridging Training And Actual Investigation Gaps In Digital Forensics,
2023
Technological Unuiversity Dublin, Ireland
Ontology-Based Case Study Management Towards Bridging Training And Actual Investigation Gaps In Digital Forensics, Hung Q. Ngo, Nhien-An Le-Khac
Articles
The training programs in digital forensics have contributed many case study models to guide digital forensic analyses. However, they only account for a small number of real cases and they are usually too abstract while actual cybercrime investigations are more diverse and complex. This gap leads to difficulties in giving immediate and straightforward actions for law enforcement during cybercrime investigations. In this paper, we propose an ontology-based knowledge map model, which is a foundation model for building a case study management system for Digital Forensic Intelligence (DFINT) and Open Source Intelligence (OSINT) in digital forensics. The main idea of this …
Group-Invariant Reinforcement Learning,
2023
San Jose State University
Group-Invariant Reinforcement Learning, Fnu Ankur
Master's Theses
Our work introduces a way to learn an optimal reinforcement learning agent accompanied by intrinsic properties of the environment. The extracted properties helps the agent to extrapolate the learning to unseen states efficiently. Out of all the various types of properties, we are intrigued towards equivariant and invariant properties, which essentially translates to symmetry. Contrary to many approaches, we do not assume the symmetry, rather learn them, making the approach agnostic to the environment and the property. The learned properties offers multiple perspective of the environment to exploit it to benefit decision making while interacting with the environment. By building …
Automatic Presentation Slide Generation Using Llms,
2023
San Jose State University
Automatic Presentation Slide Generation Using Llms, Tanya Gupta
Master's Theses
Presentation slides are widely used for conveying information in academic and professional contexts. However, manual slide creation can be time-consuming. Our research focuses on automated slide generation, specifically for scientific research papers. Automating the creation of presentation slides for scientific documents is a rather novel task and hence, there’s limited training data available and there also exists the token constraints of language models like BERT, with a maximum sequence length of 512 tokens. In this study, we fine-tune large language models, including Longformer-Encoder-Decoder (supporting sequences up to 16,834 tokens) and BIGBIRD-Pegasus (supporting sequences up to 4,096 tokens). We tackle this …
Deep Learning In Ai Medical Imaging For Stroke Diagnosis,
2023
San Jose State University
Deep Learning In Ai Medical Imaging For Stroke Diagnosis, James Mario Guzman
Master's Theses
Enhancing medical imaging stroke diagnosis applications with artificial intelligence (AI) tools to determine lesion volume, location and clinical metadata is vital toward guiding patient treatment and procedure. A major hardship in developing stroke diagnosis AI tools is the scarcity of publicly available clinical 3D stroke datasets. Through working with Johns Hopkins University, University of Michigan’s ICPSR data repository and SJSU research, we gained access to potentially the largest 3D MRI stroke dataset with clinical metadata annotated by neuroradiologists known as ICPSR 38464. With the ICPSR 38464 dataset recently being available through institutional review board (IRB) approval or exemption, we were …
Detecting The Onion Routing Traffic In Real-Time By Using Reinforcement Learning,
2023
San Jose State University
Detecting The Onion Routing Traffic In Real-Time By Using Reinforcement Learning, Dazhou Liu
Master's Theses
Anonymous networks have been popularly utilized to protect user anonymity and facilitate network security for a decade. However, such networks have been a platform for adversarial affairs and various network attacks including suspicious traffic generators. As a result, detecting anonymous network traffic is one critical task to defend a network against unpredictable attacks. Many new methods using machine learning and deep learning techniques have been proposed. However, many of them rely heavily on a vast amount of labeled data and have complicated architectures. Since network traffic always fluctuates under different network environments, those techniques may degrade in performance due to …
Intrinsic Motivation By The Principles Of Non-Linear Dynamical Systems,
2023
San Jose State University
Intrinsic Motivation By The Principles Of Non-Linear Dynamical Systems, Phu C. Nguyen
Master's Theses
The design of appropriate control rules for the stabilization of dynamical systems can require quite substantial domain knowledge. Modern AI methodologies, such as Reinforcement Learning, are often used to mitigate the need for such knowledge. However, these can be slow and often rely on at least some hand-designed reward structure, and thus human input, to be more effective. Here, we propose an alternative route to construct rewards requiring only minimal domain knowledge, essentially relying on the structure of the dynamical system itself. For this, we use truncated Lyapunov exponents as rewards to calculate the stabilizing controller from samples. Concretely, the …
Controllability-Constrained Deep Neural Network Models For Enhanced Control Of Dynamical Systems,
2023
San Jose State University
Controllability-Constrained Deep Neural Network Models For Enhanced Control Of Dynamical Systems, Suruchi Sharma
Master's Theses
Control of a dynamical system without the knowledge of dynamics is an important and challenging task. Modern machine learning approaches, such as deep neural networks (DNNs), allow for the estimation of a dynamics model from control inputs and corresponding state observation outputs. Such data-driven models are often utilized for the derivation of model-based controllers. However, in general, there are no guarantees that a model represented by DNNs will be controllable according to the formal control-theoretical meaning of controllability, which is crucial for the design of effective controllers. This often precludes the use of DNN-estimated models in applications, where formal controllability …
Comparing And Extending The Use Of Defeasible Argumentation With Quantitative Data In Real-World Contexts,
2023
Technological University Dublin
Comparing And Extending The Use Of Defeasible Argumentation With Quantitative Data In Real-World Contexts, Lucas Rizzo, Luca Longo
Articles
Dealing with uncertain, contradicting, and ambiguous information is still a central issue in Artificial Intelligence (AI). As a result, many formalisms have been proposed or adapted so as to consider non-monotonicity. A non-monotonic formalism is one that allows the retraction of previous conclusions or claims, from premises, in light of new evidence, offering some desirable flexibility when dealing with uncertainty. Among possible options, knowledge-base, non-monotonic reasoning approaches have seen their use being increased in practice. Nonetheless, only a limited number of works and researchers have performed any sort of comparison among them. This research article focuses on evaluating the inferential …
Exploring The Impact Of Noise And Degradations On Heart Sound Classification Models,
2023
Technological University Dublin
Exploring The Impact Of Noise And Degradations On Heart Sound Classification Models, Davoud Shariat Panah, Andrew Hines, Susan Mckeever
Articles
The development of data-driven heart sound classification models has been an active area of research in recent years. To develop such data-driven models in the first place, heart sound signals need to be captured using a signal acquisition device. However, it is almost impossible to capture noise-free heart sound signals due to the presence of internal and external noises in most situations. Such noises and degradations in heart sound signals can potentially reduce the accuracy of data-driven classification models. Although different techniques have been proposed in the literature to address the noise issue, how and to what extent different noise …
Comparing Poor And Favorable Outcome Prediction With Machine Learning After Mechanical Thrombectomy In Acute Ischemic Stroke,
2023
Heidelberg University Hospital, Heidelberg, Germany
Comparing Poor And Favorable Outcome Prediction With Machine Learning After Mechanical Thrombectomy In Acute Ischemic Stroke, Matthias A. Mutke, Vince I. Madai, Adam Hilbert, Esra Zihni, Arne Potreck, Charlotte S. Weyland, Markus A. Mohlenbruch, Sabine Heiland, Peter A. Ringleb, Simon Nagel, Martin Beendszus, Dietmar Frey
Articles
Outcome prediction after mechanical thrombectomy (MT) in patients with acute ischemic stroke (AIS) and large vessel occlusion (LVO) is commonly performed by focusing on favorable outcome (modified Rankin Scale, mRS 0–2) after 3 months but poor outcome representing severe disability and mortality (mRS 5 and 6) might be of equal importance for clinical decision-making.
Corrigendum: Human Mental Workload: A Survey And A Novel Inclusive Definition,
2023
Technological University Dublin
Corrigendum: Human Mental Workload: A Survey And A Novel Inclusive Definition, Luca Longo, Christopher D. Wickens, Gabriella Hancock, P. A. Hancock
Articles
In the published article, the name of Gabriella Hancock was incorrectly written as “Gabriela M. Hancock.” The correct name is “Gabriella Hancock.” In the published article, there was also an error in the author list as published. Gabriella Hancock was listed as the last author, but should have been listed as third author. P. A. Hancock was listed as third author but should be listed as the last author. The corrected author list appears below. Luca Longo1, Christopher D.Wickens, Gabriella Hancock and P. A. Hancock. The authors apologize for this error and state that this does not change the scientific …
Towards Automated Weed Detection Through Two-Stage Semantic Segmentation Of Tobacco And Weed Pixels In Aerial Imagery,
2023
a National University of Sciences and Technology H-12, Islamabad, Pakistan
Towards Automated Weed Detection Through Two-Stage Semantic Segmentation Of Tobacco And Weed Pixels In Aerial Imagery, S. Imran Moazzam, Umar S. Khan, Waqar Qureshi, Tahir Nawaz, Faraz Kunwar
Articles
In precision farming, weed detection is required for precise weedicide application, and the detection of tobacco crops is necessary for pesticide application on tobacco leaves. Automated accurate detection of tobacco and weeds through aerial visual cues holds promise. Precise weed detection in crop field imagery can be treated as a semantic segmentation problem. Many image processing, classical machine learning, and deep learning-based approaches have been devised in the past, out of which deep learning-based techniques promise better accuracies for semantic segmentation, i.e., pixel-level classification. We present a new method that improves the precision of pixel-level inter-class classification of the crop …
