Publications
Our lab is dedicated to advancing knowledge through rigorous research and impactful scholarship. Below is a curated list of our peer-reviewed journal articles, conference papers, preprints, and other scholarly contributions.
2026
Real-time American Sign Language Subtitle Generation using NLP Models for Enhanced Accessibility
The accessibility of digital media for Deaf and Hard of Hearing (DHH) individuals continues to present significant challenges, as traditional subtitles often fail to adequately capture the nuances of American Sign Language (ASL). This research introduces a real-time ASL subtitle generation system le...
The accessibility of digital media for Deaf and Hard of Hearing (DHH) individuals continues to present significant challenges, as traditional subtitles often fail to adequately capture the nuances of American Sign Language (ASL). This research introduces a real-time ASL subtitle generation system leveraging advanced speech-to-text and computer vision techniques. We utilized OpenAI's Whisper ASR (Automatic Speech Recognition) system to achieve rapid and accurate real-time speech transcription, effectively converting spoken language into textual format. Subsequently, ASL translations were generated using prerecorded ASL videos from Handspeak.com, processed through Google's MediaPipe Holistic solution to extract detailed facial, pose, and hand landmark coordinates. A stick-figure avatar was developed to visually represent these landmarks clearly and dynamically, providing users with an intuitive understanding of ASL gestures. The real-time system integrates microphone input and live webcam feeds, offering immediate visual translations with minimal latency. While current model accuracy and vocabulary coverage face limitations due to dataset constraints, initial implementations demonstrate practical feasibility and substantial potential for enhancing digital inclusivity for the DHH community.
HEX: Agentic Framework for Closed-domain Web Tables Generation
Existing WebListQA frameworks mostly treat list-based web queries as an entity extraction task, failing when answers require multi-attribute information of entities. We formalize this limitation as comprehensive retrieval, a high-recall task of synthesizing structured tables from web sources. While ...
Existing WebListQA frameworks mostly treat list-based web queries as an entity extraction task, failing when answers require multi-attribute information of entities. We formalize this limitation as comprehensive retrieval, a high-recall task of synthesizing structured tables from web sources. While recent agentic retrieval frameworks can autonomously browse the web to retrieve information, they lack effective query reformulation and structured boundaries necessary to aggregate web entities into a comprehensive table. To address these limitations, we introduce HEX, a six-state agentic framework that leverages LLM-based decision making within state-machine boundaries. HEX ensures exhaustive retrieval of entities and their attributes to generate a comprehensive web table. Evaluating HEX against state-of-the-art baselines across three healthcare domain tasks demonstrate that HEX improves entity recall from 10\
Does Context Compression Preserve Refusal Alignment?
Context compression reduces inference cost by encoding inputs into compact representations while preserving semantic content. An open question is whether semantic preservation alone is sufficient to maintain downstream behaviours such as refusal alignment. We investigate this question and find that ...
Context compression reduces inference cost by encoding inputs into compact representations while preserving semantic content. An open question is whether semantic preservation alone is sufficient to maintain downstream behaviours such as refusal alignment. We investigate this question and find that encoder-based compression systematically weakens refusal behaviour in instruction-tuned language models, despite high reconstruction fidelity. This effect persists across model families and compression architectures. Mechanistic analysis shows that compression attenuates activation along the decoder’s learned refusal direction. We further explore , a lightweight inference-time intervention that restores refusal rates to near-baseline levels without retraining and operates entirely in compressed representation space. These results demonstrate that semantic preservation does not guarantee behavioural preservation under compression, highlighting the need to explicitly preserve alignment-relevant features in compression-aware systems.
Temporal XAI for Model Explanation Drift Monitoring: Early Warning of Population Mental Health Volatility
TASR: A Trustworthy LLM-based Framework for TCFD-Aligned Sustainability Report Analysis
Reliable and transparent assessment of environmental, social, and governance (ESG) disclosures is critical for sustainable finance, regulatory oversight, and risk-aware decision-making. However, existing sustainability reporting evaluations rely on costly manual reviews or third-party ratings, which...
Reliable and transparent assessment of environmental, social, and governance (ESG) disclosures is critical for sustainable finance, regulatory oversight, and risk-aware decision-making. However, existing sustainability reporting evaluations rely on costly manual reviews or third-party ratings, which limit reproducibility. This work proposes a trustworthy large language model (LLM)-based framework for automated sustainability report analysis aligned with the Task Force on Climate-related Financial Disclosures (TCFD). We propose TASR (Trustworthy Analysis for Sustainability Report), a three-stage framework for TCFD-aligned sustainability report analysis that integrates LLM-based scoring, benchmarking against third-party ESG ratings, and downstream predictive modeling. Experiments on 100 sustainability reports from U.S. oil, gas, and mining companies demonstrate strong alignment with Bloomberg Environmental Disclosure scores and high score stability across repeated evaluations. Furthermore, predictive models trained on the LLM-generated TCFD scores achieve meaningful predictive performance in forecasting disclosure benchmarks, highlighting their practical utility for sustainability rating. The results suggest that LLM-based TCFD scoring offers a potentially scalable and transparent alternative for sustainability disclosure assessment.
Elements of virtual navigators for health and social services: Protocol for a scoping review
Patient navigation is a critical component of health care delivery, facilitating connections with appropriate services. A new era of virtual navigators are being developed that can be accessed through the web or smartphone application. However, it is unknown what features make virtual navigators acc...
Patient navigation is a critical component of health care delivery, facilitating connections with appropriate services. A new era of virtual navigators are being developed that can be accessed through the web or smartphone application. However, it is unknown what features make virtual navigators accessible and reliable. The objective of this scoping review is to understand the current landscape of existing virtual navigation systems. In this review we will determine the features of these systems and barriers to accessibility they've identified. This review will follow the guidelines for scoping reviews outlined by the Joanna Briggs Institute (JBI) methodology. A search strategy will be used to locate both published and unpublished literature.
Context-based Ontology Modelling for Database: Enabling ChatGPT for Semantic Database Management
This research paper explores the use of ChatGPT in database management. ChatGPT, an AI-powered chatbot, has limitations in performing tasks related to database management due to the lack of standardized vocabulary and grammar for representing database semantics. To address this limitation, the paper...
This research paper explores the use of ChatGPT in database management. ChatGPT, an AI-powered chatbot, has limitations in performing tasks related to database management due to the lack of standardized vocabulary and grammar for representing database semantics. To address this limitation, the paper proposes a solution that involves developing a set of syntaxes that can represent database semantic meaning in a way that ChatGPT can understand and use to perform database management tasks.
Support-Aware Point Cloud Upsampling
Point cloud upsampling methods are commonly evaluated using geometric similarity metrics such as Chamfer and Hausdorff distance, despite these objectives not necessarily reflecting whether added points meaningfully contribute to downstream reconstruction. This work investigates point cloud upsamplin...
Point cloud upsampling methods are commonly evaluated using geometric similarity metrics such as Chamfer and Hausdorff distance, despite these objectives not necessarily reflecting whether added points meaningfully contribute to downstream reconstruction. This work investigates point cloud upsampling as a support-allocation problem through a support-aware implicit projection framework that emphasizes ownership specialization, reconstruction-oriented support placement, and usefulness-aware evaluation rather than uniform density generation alone.
Argus: A Game-Agnostic Accessibility Overlay for Visually Impaired Players
More than 2.2 billion people globally live with near or distance vision impairment, yet video game accessibility depends almost entirely on per-game effort: studios that do not invest in built-in features leave their titles permanently unreachable, and community-made accessibility mods, though valua...
More than 2.2 billion people globally live with near or distance vision impairment, yet video game accessibility depends almost entirely on per-game effort: studios that do not invest in built-in features leave their titles permanently unreachable, and community-made accessibility mods, though valuable, are confined to games that expose a modding API. We present Argus, a game-agnostic accessibility overlay that delivers real-time spoken narration for any video game without game-side modification or developer cooperation, operating solely on the rendered screen output. Argus captures the live game window, segments UI regions using open-vocabulary prompts (SAM~3), extracts text via OCR, and synthesises context-aware guidance through a local LLM and neural TTS engine---a pipeline that generalises across art styles without per-title training data. We demonstrate Argus on tutorial gameplay from and , two visually and mechanically distinct titles, confirming that the same prompt vocabulary transfers across both rendering styles without reconfiguration.
2025
An Interactive AI Solution for Market Research and Report Generation
In the domain of business intelligence, the ability to efficiently synthesize and analyze extensive secondary market research from a plethora of sources represents a critical challenge. This paper outlines the development of an advanced analytical tool that utilizes Large Language Models (LLMs) and ...
In the domain of business intelligence, the ability to efficiently synthesize and analyze extensive secondary market research from a plethora of sources represents a critical challenge. This paper outlines the development of an advanced analytical tool that utilizes Large Language Models (LLMs) and a sophisticated Graphical User Interface (GUI) to streamline the synthesis and analysis of extensive secondary market research. Designed to gather and interpret data from various sources such as news articles, product releases, and professional profiles, the tool provides tailored insights to answer specific business questions. It enables users to interact directly with AI-generated outputs to refine and customize the analysis, ensuring both relevance and accuracy. The tool uniquely supports the creation of detailed market research reports and concise PowerPoint presentations to meet the diverse needs of stakeholders. The paper also discusses the integration of a user feedback loop, which enhances the system’s learning capabilities for continuous performance improvement. The outcome of the paper demonstrates that this tool significantly boosts efficiency, precision, and stakeholder engagement in market research analysis, marking a new phase in human-AI collaboration in business intelligence. Index Terms-Large Language Model, Market Research, Interactive Analytics, Competitive Intelligence.
Dynamic Web Page Modification for Accessibility Using AI and Large Language Models
Web accessibility remains a critical challenge, especially for individuals with disabilities, as most web content is designed without comprehensive adherence to accessibility standards. Current accessibility solutions are limited, relying heavily on the voluntary incorporation of specific design and...
Web accessibility remains a critical challenge, especially for individuals with disabilities, as most web content is designed without comprehensive adherence to accessibility standards. Current accessibility solutions are limited, relying heavily on the voluntary incorporation of specific design and coding standards during web page development. This paper introduces a novel application of artificial intelligence, specifically through the use of Large Language Models (LLMs), to address these challenges. By enabling LLMs to analyze and adaptively modify web page source code, our approach allows for the dynamic standardization of web pages to incorporate user-controllable accessibility features such as text resizing, color contrast adjustments, and font changes. Through the Adaptive User Interface Framework (AUIF), our system adapts digital content in real-time, responding to individual user behaviors and preferences, thereby setting a new standard for accessible user experiences across various digital platforms.
Navigational Assistance for the Blind in Complex Indoor Spaces Using a Vision-Enabled Large Language Model
This study introduces an innovative implementation of a Large Language Model (LLM) that leverages both vision and natural language processing to enhance navigation for individuals who are blind. Unlike traditional methods that rely on pre-existing maps or environmental reconstruction using sensors l...
This study introduces an innovative implementation of a Large Language Model (LLM) that leverages both vision and natural language processing to enhance navigation for individuals who are blind. Unlike traditional methods that rely on pre-existing maps or environmental reconstruction using sensors like LiDAR, our approach requires no prior environmental data and instead utilizes real-time visual cues similar to human navigation strategies. This novel methodology allows the model to dynamically interpret and verbalize complex indoor environments, providing blind users with descriptive audio cues that effectively convey the spatial layout and pertinent features of their surroundings. Conducted in a hospital setting, our experiments demonstrated that this approach significantly improves GPT4-V’s navigation capabilities and offers real-time, contextually relevant guidance, thereby enhancing the independence and safety of blind individuals navigating complex spaces. This research contributes to the understanding of AI’s capabilities in real-world applications and opens new avenues for the deployment of language models in complex, dynamic environments.
Integrating Ethical AI Tools into Educational Practices for Enhancing Academic Integrity
As large language models (LLMs) become increasingly integrated into educational settings, concerns about academic integrity, ethical usage, and student engagement are becoming more prominent. While these AI tools can effectively provide personalized learning experiences and support diverse student n...
As large language models (LLMs) become increasingly integrated into educational settings, concerns about academic integrity, ethical usage, and student engagement are becoming more prominent. While these AI tools can effectively provide personalized learning experiences and support diverse student needs, they also risk overreliance and promote unethical academic practices if used without appropriate safeguards. This paper presents a novel approach that integrates an LLM-based assistant directly into a learning management system (LMS) with carefully designed constraints to encourage active learning, reduce misuse, and preserve academic integrity. We establish core design principles to address the challenges associated with LLMs in education and provide a detailed description of our system’s architecture. Additionally, we conduct a pilot study to assess the tool’s impact on student learning and gather feedback for further improvements. A prototype of the tool is publicly available on Github.
A novel digital twins-driven mutual trust framework for human--robot collaborations
Trust plays an important role and significantly influences human–robot collaborations (HRC). However, most previous research on trust only emphasizes the human attitude toward robots. There needs more understanding of human uncertainties that may also cause disruptions of trust in collaborations. Th...
Trust plays an important role and significantly influences human–robot collaborations (HRC). However, most previous research on trust only emphasizes the human attitude toward robots. There needs more understanding of human uncertainties that may also cause disruptions of trust in collaborations. This paper presents a novel mutual trust framework to provide a relatable vision for future development in HRC from an integrated perspective via the integration of human and robotic digital twins. More specifically, a comprehensive review of current trust research in HRC is first provided, including trust factors and state-of-the-art trust models. Second, a novel human–robot mutual trust framework based on 5-layer digital twins models is introduced. The mutual trust framework highlights the interactions amongst modules of artificial intelligence, simulation, and operation, which can provide wide services in HRC (e.g., task allocation and motion planning). A case study of solving a path planning problem is exemplified to evaluate the performance of the proposed mutual trust framework. Compared with singular trust models, the proposed framework enables robotic systems with real-time response and adaptation to human behavior. Some limitations and future work of the mutual trust framework are elaborated in the end.
Leveraging LASSO-Based Methodologies for Enhanced SNP Analysis in Plant Genomes
Abstract Summary Genome-wide association studies (GWAS) have been widely used to reveal the associations between genetic variations and phenotypes in a population of individuals. However, they have been criticized for missing important genetic markers usually due to the fact that the data may not fi...
Abstract Summary Genome-wide association studies (GWAS) have been widely used to reveal the associations between genetic variations and phenotypes in a population of individuals. However, they have been criticized for missing important genetic markers usually due to the fact that the data may not fit the statistical models well. In this study, we address the challenge of identifying significant single nucleotide polymorphisms (SNPs) in GWAS by harnessing the capabilities of two sophisticated regression models, BIGLASSO and AUTALASSO. They are both variants of the least absolute shrinkage and selection operator (LASSO). Our research contributes to the field of genomics through detailed comparative analysis of Arabidopsis thaliana, revealing how each method specializes in uncovering SNPs for different trait types. Our findings indicate that BIGLASSO shows stronger alignment with GWAS results, particularly excelling in the analysis of binary traits, even when these are derived from categorical phenotypes. AUTALASSO could be effective for quantitative traits and complement GWAS. We demonstrate that these LASSO-based methods can significantly enhance the identification of genetic markers, offering a potent complement to traditional GWAS approaches. Our findings not only bridge the gap between statistical and machine learning methodologies in genetic studies but also provide a practical framework for researchers seeking to validate reported SNPs or explore new genomic regions for trait association. This work stands as a pivotal step toward the integration of advanced computational techniques in genomics, paving the way for more precise and comprehensive genetic analyses. Availability and implementation Key results from the paper are available at the https://github.com/DongdongHou006/LASSO-SNP. The program was implementated using Python and R, and was tested using the Digital Research Alliance of Canada.
Enhancing cancer subtype classification through convolutional neural networks: a deepinsight analysis of TCGA gene expression data
This study investigates the adaptation of DeepInsight for cancer subtype classification using high-dimensional gene expression data. It was initially designed for non-image data and has been adapted for cancer classification. We compared DeepInsight's performance with traditional models, including s...
This study investigates the adaptation of DeepInsight for cancer subtype classification using high-dimensional gene expression data. It was initially designed for non-image data and has been adapted for cancer classification. We compared DeepInsight's performance with traditional models, including support vector machine, LightGBM, neural networks, and decision trees, both with and without synthetic Minority Oversampling Technique. DeepInsight consistently outperformed these models in terms of F1 score across breast, lung, and colon cancer datasets. It effectively captures complex gene relationships, addressing multi-class classification challenges. A key innovation is modifying a multi-class feature selection method using aggregated class activation maps, which capture top genes across subtypes. Additionally, we identified top genes using several models and found that several key genes were consistently recognized across multiple methods, such as MLH1 in colon cancer. Gene Ontology analysis of these critical genes revealed their involvement in significant biological processes. DeepInsight Cancer Subtype Classification Convolutional Neural Networks Gene Expression Data Comparative Study
Third-Party Privacy Data Leak Analysis on Ontario Hospital Websites
This study investigates the privacy concerns associated with third-party data leaks on hospital websites in Ontario, Canada. The study employs both manual and automated analytics to investigate the privacy practices of 135 hospital websites in Ontario. This dual approach entails a thorough evaluatio...
This study investigates the privacy concerns associated with third-party data leaks on hospital websites in Ontario, Canada. The study employs both manual and automated analytics to investigate the privacy practices of 135 hospital websites in Ontario. This dual approach entails a thorough evaluation of websites while gathering and evaluating datasharing policies. The findings show substantial variation in datasharing procedures, with some hospital websites disclosing user data to third parties without explicit agreement, posing serious privacy issues. The study uncovers trends and abnormalities in data transfers, emphasizing the need for regulatory measures and enforcement to secure patient information. Furthermore, the findings highlight the critical need for strong privacy safeguards and regulatory frameworks in the healthcare industry. The study also makes practical recommendations for improving data privacy on hospital websites.
LESA: LLM-based Search Assistant for Healthcare Comprehensive Information Retrieval
Comprehensive Information Retrieval (CIR) is a challenging task in information-intensive domains like healthcare. While RAG excels at retrieving concise and relevant answers, it falls short of delivering the structured and comprehensive overviews that CIR requires. To address this, we introduce LESA...
Comprehensive Information Retrieval (CIR) is a challenging task in information-intensive domains like healthcare. While RAG excels at retrieving concise and relevant answers, it falls short of delivering the structured and comprehensive overviews that CIR requires. To address this, we introduce LESA, a novel model that leverages a large language model for query reformulation to orchestrate a RAG-based search agent. LESA reformulates queries, performs multi-step retrieval, information extraction, and synthesizes the results into a search transcript. We evaluate LESA against RAG and other models built in this study across different healthcare domains. Our results show that LESA significantly outperforms all baselines in terms of recall and utility. Our findings suggest that LLM-driven session-level orchestration is a promising direction for information-intensive domains.
CHARM: Leveraging Reused Medical Knowledge Graphs and LLMs for Community Health Resource Recommendation
Access to appropriate community and healthcare resources is critical for addressing both medical and non-medical determinants of health. However, developing reliable recommendation systems for such resources is challenging due to the scarcity of domain-specific structured knowledge. Although knowled...
Access to appropriate community and healthcare resources is critical for addressing both medical and non-medical determinants of health. However, developing reliable recommendation systems for such resources is challenging due to the scarcity of domain-specific structured knowledge. Although knowledge graphs (KGs) are a promising solution for enriching decision-making in recommendation systems, constructing a domain-specific KG typically requires significant manual effort and curated datasets, resources that are not always available in community health contexts.In this study, we explore the feasibility of repurposing an existing knowledge graph originally designed for precision medicine, PrimeKG, to support resource recommendation in a broader community health context. We introduce CHARM (Community Health Assistance via Reused Medical knowledge graphs), a recommendation system framework that integrates Large Language Models (LLMs) with adapted KGs. CHARM utilizes LLMs to interpret unstructured physician notes and extract patient needs, while the repurposed knowledge graph supplements contextual medical and social insights to enhance recommendation accuracy.We examine four strategies for incorporating knowledge graphs into the LLM-driven pipeline and evaluate their effectiveness using 58 diverse patient scenarios. Our results demonstrate that integrating the knowledge graph between the keyword extraction and problem identification stages (the CHARM strategy) yields the highest performance in both keyword and resource recommendation tasks. Notably, the approach performs well even for non-biomedical scenarios, validating the transferability of the precision medicine KG to broader healthcare domains.This work contributes a scalable, hybrid methodology for community resource recommendation that reduces reliance on domain-specific KGs. It also provides empirical evidence supporting the cross-domain reuse of biomedical KGs to assist in addressing the social determinants of health.
ODLP: Scalable Multimodal Information Retrieval Using Visual and Structural Integration
The effective application of Large Language Models (LLMs) in domain-specific settings (e.g., e-commerce, finance, science) hinges on their ability to access and reason over reliable structured knowledge. However, extracting structured knowledge from e-commerce webpages presents significant challenge...
The effective application of Large Language Models (LLMs) in domain-specific settings (e.g., e-commerce, finance, science) hinges on their ability to access and reason over reliable structured knowledge. However, extracting structured knowledge from e-commerce webpages presents significant challenges. First, webpage content is typically represented by HTML and CSS, which, when directly inputted into Large Language Models (LLMs), frequently exceeds token limitations. Second, effectively transforming unstructured web data into structured information, such as extracting detailed lists of product descriptions, remains problematic and necessitates sophisticated parsing techniques. This paper investigates the integration of unstructured knowledge, exemplified by HTML, with other modalities (like visual representation) to derive structured knowledge tailored for e-commerce domain tasks. This method enables accurate, context-aware extraction and alignment of item information (e.g., product attributes, pricing), overcoming the limitations of methods reliant solely on unstructured text or inconsistent tags. Experimental evaluations across 31 shopping websites with more than 1,200 products validate the effectiveness of this structured knowledge integration, achieving 100 recall/precision and demonstrating robustness. ODLP significantly outperforms LLM-based tools like ZeroX, showcasing the power of combining multimodal data with LLM reasoning for domain-specific problems. This work provides a reliable method for processing extensive amount of unstructured web information.
Breaking the linear barrier: A multi-modal LLM-based system for navigating complex web content
Visually impaired users still face fundamental obstacles when interacting with complex, dynamic websites. Conventional screen readers expose pages in a strict linear order, offer little semantic context for visual media, and provide limited context regarding the page content. This paper introduces a...
Visually impaired users still face fundamental obstacles when interacting with complex, dynamic websites. Conventional screen readers expose pages in a strict linear order, offer little semantic context for visual media, and provide limited context regarding the page content. This paper introduces a multi-modal accessibility framework combining Large Language Models (LLMs), Computer Vision, and dynamic DOM manipulation to significantly enhance semantic clarity, non-linear navigation, and interaction richness. By interpreting visual and textual web content contextually and adapting it into an intuitive, conversationally navigable interface, our method provides a foundation for visually impaired users to interact effectively with previously inaccessible or challenging digital experiences.The deployment of a functional prototype on a modern web browser illustrates the capability of the proposed system to interact with diverse websites and tasks. The research team selected Canada’s most frequented websites to assess the system’s efficacy in enhancing contextual understanding of the page content and enabling navigation through pages and actions via a chat-driven interface. A comprehensive demonstration was executed using a prominent ticketing site, which facilitated users in obtaining a deeper understanding of the page while guiding them towards the successful purchase of concert tickets. By illustrating how vision language and LLM reasoning can be coupled with low-level browser control, this work lays the groundwork for future efforts in performance optimization, large-scale evaluation, and personalization across diverse web contexts.
Cogniroot Edge: A quest for a fair AI grader
This paper proposes an AI-based grading software to ensure fair and high-quality grading for large classes. Fair and high-quality grading is a fundamental necessity in an education system; however, it can be quite challenging for large classes. Due to manpower limitations and the complexity of gradi...
This paper proposes an AI-based grading software to ensure fair and high-quality grading for large classes. Fair and high-quality grading is a fundamental necessity in an education system; however, it can be quite challenging for large classes. Due to manpower limitations and the complexity of grading, multiple-choice questions (MCQs) are often the preferred approach for assessing the performance of large classes. MCQs allow automated grading, minimizing manpower requirements and grading errors. However, MCQ-based approaches limit the quality of the evaluation process since they do not accommodate other forms of assessment, such as short-answer questions, essays, and mathematical problems. To address this challenge, this research work developed an AI-based grading method capable of automatically assessing short answers, essays, and math problems. This research conducted extensive usability testing on the developed AI grader. The results indicate that the AI grader has significant potential to enhance the evaluation process for large classes; however, there is room for improvement in its performance which have been identified and highlighted in this paper for further research in this domain.
Predicting T cell receptor specificity with graph attention networks
T cells play a pivotal role in adaptive immunity by recognizing a vast array of antigens through their T cell receptors (TCRs). Advances in high-throughput sequencing have enabled unprecedented profiling of TCR repertoires, enabling the computationally prediction of TCR-antigen associations. Existin...
T cells play a pivotal role in adaptive immunity by recognizing a vast array of antigens through their T cell receptors (TCRs). Advances in high-throughput sequencing have enabled unprecedented profiling of TCR repertoires, enabling the computationally prediction of TCR-antigen associations. Existing machine learning approaches, including deep learning and clustering algorithms, have shown promise in identifying patterns within TCR sequences, particularly through features such as complementarity-determining region 3 (CDR3) sequences and V(D)J gene usage. However, many existing models are constrained by their architectural limitations, and often relying on feature concatenation to achieve multi-modal analysis. In this study, we proposed an algorithm that integrates deep learning techniques with immune repertoire data to enhance TCR specificity prediction. By leveraging language models and graph attention networks, our framework systematically identifies patterns that correlate with antigen recognition, improving predictive accuracy and enhancing multi-modal analysis of TCR sequencing data. Experiment results obtained from two public datasets showed that our method can accurately predict TCR-epitope associations. Our study also underscores the importance of integrating biomolecular networks and advanced language models to uncover fundamental principles of immune responses and translate them into clinically relevant applications. Graph attention networks, T cell receptor specificity, Heterogeneous network, BERT
Prototyping with generative AI in an extended reality course: From ideation to implementation
Cost-effective predictive modeling for student mental health using readily-available data
The mental health of post-secondary students is a critical public health issue, with alarming rates of psychological distress, suicidal thoughts, and behaviors on university and college campuses. Predictive modeling can be utilized for the analysis of student mental health to better understand curre...
The mental health of post-secondary students is a critical public health issue, with alarming rates of psychological distress, suicidal thoughts, and behaviors on university and college campuses. Predictive modeling can be utilized for the analysis of student mental health to better understand current students' mental state. These predictions are often made using large surveys collected from students or participants to use the survey questions as features and then predict based on a target question related to mental health. The expensive nature of collecting data this way can be prohibitive for some institutions, and due to the scale and potential data processing required, the predictions made using those data could be too late for any proactive approaches to tackle the mental health of students. To address this, it is worth investigating the predictive performance of readily available data to predict student mental health as a means of accurately representing an institution's student body. In this paper, we show that readily-available data can be used to predict mental health with competitive accuracy compared to other experiments done in the literature that utilize more expensively collected data with neural network models.
Creation of visualizations with a multi-agent LLM approach
Data visualization is a critical tool for interpreting complex information, yet it often remains inaccessible to those without extensive technical and analytical skills. This study introduces a novel multi-agent system leveraging a large language model (LLM) to democratize the process of creating hi...
Data visualization is a critical tool for interpreting complex information, yet it often remains inaccessible to those without extensive technical and analytical skills. This study introduces a novel multi-agent system leveraging a large language model (LLM) to democratize the process of creating high-quality visualizations. By automating the stages of planning, coding, and interpretation, the system empowers users with diverse backgrounds to generate accurate and meaningful visual representations of data. Our approach employs multiple specialized agents, each focusing on different aspects of the visualization workflow, thereby enhancing the overall quality through collaborative problem-solving and contextual communication. The iterative refinement phase ensures that the visualizations meet the initial objectives and data characteristics, thus improving accuracy and relevance. This study's modular design allows for scalability and adaptability to various data types and visualization needs, ensuring the system remains current with emerging tools and frameworks. By lowering the barriers to effective data visualization, our system supports broader data-driven decision-making across various domains, fostering more inclusive and impactful data analysis practices. Validation on two public datasets demonstrates that our multi-agent framework generates visualizations that achieve comparable or superior quality metrics when benchmarked against human expert analysis.
Intra-Layer Recurrence in Transformers for Language Modeling
Transformer models have established new benchmarks in natural language processing; however, their increasing depth results in substantial growth in parameter counts. While existing recurrent transformer methods address this issue by reprocessing layers multiple times, they often apply recurrence ind...
Transformer models have established new benchmarks in natural language processing; however, their increasing depth results in substantial growth in parameter counts. While existing recurrent transformer methods address this issue by reprocessing layers multiple times, they often apply recurrence indiscriminately across entire blocks of layers. In this work, we investigate Intra-Layer Recurrence (ILR), a more targeted approach that applies recurrence selectively to individual layers within a single forward pass. Our experiments show that allocating more iterations to earlier layers yields optimal results. These findings suggest that ILR offers a promising direction for optimizing recurrent structures in transformer architectures.
LLM-powered SQL querying: Transforming natural language into database insights
In modern database management and business intelligence, bridging the gap between natural language and structured query execution remains a crucial challenge. This paper presents a system that leverages Large Language Models (LLMs) to enable natural language interaction with databases, allowing user...
In modern database management and business intelligence, bridging the gap between natural language and structured query execution remains a crucial challenge. This paper presents a system that leverages Large Language Models (LLMs) to enable natural language interaction with databases, allowing users to query a locally hosted database without requiring SQL expertise. The system consists of a user interface that accepts natural language prompts, translates them into SQL queries using an LLM, executes the queries on the database, and returns the responses in JSON format. By eliminating the need for manual query writing, this approach enhances accessibility, efficiency, and usability for nontechnical users. Furthermore, the system incorporates a feedback loop to refine query generation accuracy over time. The results demonstrate improved query execution efficiency and usability, marking a significant step in AI-driven database interaction.
Assessing privacy practices on Ontario municipal websites
The sharing of personal data on government websites is a major concern of daily users. The existing regulations do not allow for the collection and sharing of personal data. This study investigates the privacy practices of 444 municipal websites across Ontario, Canada, focusing on compliance with re...
The sharing of personal data on government websites is a major concern of daily users. The existing regulations do not allow for the collection and sharing of personal data. This study investigates the privacy practices of 444 municipal websites across Ontario, Canada, focusing on compliance with relevant data protection regulations and the extent of third-party data sharing. In particular, we examine the issues in line with Canadian standards, the Personal Information Protection and Electronic Documents Act (PIPEDA), the Canadian Privacy Act (CPA), and the Municipal Freedom of Information and Protection of Privacy Act (MFIPPA). We perform network traffic analysis, and apply a combination of privacy policies. Our findings uncover substantial gaps in privacy practices, including insufficient transparency, inadequate user-consent mechanisms, and pervasive third-party data sharing. The results of our study highlight an urgent need for the enhancement of privacy measures on government municipal websites to protect the personal data of users and for the implementation of practices that comply with local and international privacy laws, such as PIPEDA, CPA, MFIPPA, and the General Data Protection Regulation (GDPR). The study provides actionable recommendations aimed at strengthening data protection and restoring public trust in digital municipal services.
Privacy-preserving machine learning for mental health prediction using homomorphic encryption
Student mental health issues, such as stress, anxiety, and depression, are increasingly prevalent in academic institutions, significantly affecting well-being and academic performance. Recent machine learning (ML)-based systems have demonstrated promise in predicting mental health conditions using s...
Student mental health issues, such as stress, anxiety, and depression, are increasingly prevalent in academic institutions, significantly affecting well-being and academic performance. Recent machine learning (ML)-based systems have demonstrated promise in predicting mental health conditions using survey data, but these approaches often process sensitive information in plaintext, risking privacy breaches or relying on centralized data storage vulnerable to leaks. Homomorphic encryption (HE) has been proposed for secure ML, but existing implementations either focus on simpler datasets (e.g., numerical/IoT data) or incur impractical computational overhead (e.g., high RAM usage or prolonged training times) for real-world mental health applications. To address these gaps, we introduce a privacy-preserving predictive model for student mental health using logistic regression trained directly on encrypted data via the TenSEAL library. Our work uniquely combines a leveled fully homomorphic encryption (FHE) scheme to ensure end-to-end confidentiality, replacing the standard sigmoid with a quadratic approximation for homomorphic compatibility. We also perform a comprehensive efficiency analysis that evaluates RAM usage and training time across polynomial-modulus degrees to balance security and practicality, a trade-off underexplored in prior HEbased mental health studies. Experimental results show that our encrypted model achieves 84% accuracy (vs. 96% unencrypted) with minimal performance loss, while benchmarks demonstrate scalable resource consumption. This work advances the feasibility of implementing FHE in sensitive domains such as mental health, offering a rigorous template for privacy-preserving ML without compromising predictive utility.
Towards Efficient Keyframe Selection for News Video Captioning
Vision-language models are increasingly used for video captioning, but their reliance on uniform frame sampling introduces inefficiencies. Uniform sampling risks overlooking informative frames rich in textual or structural cues while redundantly processing visually similar frames, leading to inflate...
Vision-language models are increasingly used for video captioning, but their reliance on uniform frame sampling introduces inefficiencies. Uniform sampling risks overlooking informative frames rich in textual or structural cues while redundantly processing visually similar frames, leading to inflated computational costs and degraded caption quality. To address this, we propose a layout-aware keyframe selection method that integrates shot segmentation with layout detection to identify frames containing the most semantically informative visual and textual elements. By prioritizing event-rich frames, our approach reduces redundancy while preserving critical contextual information for downstream captioning. On a benchmark of English-language news videos, our method achieved an event matching F1 score of 0.961, a 14\
Strengthening Computer Science Education Through XR: Opportunities, Challenges, and Pedagogical Pathways
2024
Beyond Traditional Teams: Using ChatGPT to Simulate Project Management Dynamics and Software Development in Online Higher Education
In an era characterized by asynchronous online education, effectively teaching project management poses unique challenges. This paper presents a pioneering approach wherein ChatGPT, an advanced conversational AI, is integrated into an undergraduate Systems Analysis & Analytics course to simulate tea...
In an era characterized by asynchronous online education, effectively teaching project management poses unique challenges. This paper presents a pioneering approach wherein ChatGPT, an advanced conversational AI, is integrated into an undergraduate Systems Analysis & Analytics course to simulate team dynamics. Beyond merely emulating interactions, ChatGPT assumes distinct roles-from UI/UX Designers to backend developers-enabling students to experience project management without the logistical complications of coordinating with real team members. We delve further into an exploratory realm, evaluating the feasibility of students collaboratively developing actual software, a Chrome extension in this case, in tandem with ChatGPT. Preliminary feedback suggests an enriching and consistent educational experience, emphasizing the transformative potential of AI-driven agents in pedagogical settings. The paper sets the stage for a broader discussion on the future of education, inviting readers to consider the implications of AI not just as a tool, but as a collaborator in teaching, learning and development.
Bridging the Accessibility Gap in Online Shopping with AI-Driven Solutions
Despite advancements in assistive technologies, blind and low vision (BLV) individuals continue to face significant challenges in online shopping, particularly with interpreting visual content and comparing products. Existing tools often lack seamless integration and intuitive interaction, hindering...
Despite advancements in assistive technologies, blind and low vision (BLV) individuals continue to face significant challenges in online shopping, particularly with interpreting visual content and comparing products. Existing tools often lack seamless integration and intuitive interaction, hindering an equitable shopping experience. This paper introduces Shop Sight, a Chrome browser extension designed to enhance online shopping accessibility for BLV users. Shop Sight leverages artificial intelligence (AI) and voice-activated capabilities to generate context-rich product image descriptions and to facilitate simplified product comparisons through voice commands. The development and evaluation of Shop Sight demonstrate its potential to bridge the gap between current AI capabilities and the real-world needs of BLV users in e-commerce. By providing AI-generated image descriptions and voice-activated product comparisons, Shop Sight empowers BLV individuals with greater independence and a more personalized, efficient online shopping experience. Future work will focus on refining the tool's features, incorporating user feedback, and addressing identified limitations to further enhance its usability and effectiveness.
Enhancing healthcare user interfaces through large language models within the adaptive user interface framework
In the pursuit of enhancing digital user experiences within healthcare, this research investigates the novel application of Large Language Models (LLMs) in the Adaptive User Interface Framework (AUIF). This framework aims to redefine user interaction by providing real-time, personalized interface ad...
In the pursuit of enhancing digital user experiences within healthcare, this research investigates the novel application of Large Language Models (LLMs) in the Adaptive User Interface Framework (AUIF). This framework aims to redefine user interaction by providing real-time, personalized interface adjustments. By systematically applying LLMs to user interface enhancement, the AUIF addresses the static nature of current digital health platforms, offering a dynamic and adaptive alternative that responds to individual user behaviors and preferences. This study explores the pioneering integration of LLMs for user experience (UX) improvement recommendations and real-time HyperText Markup Language (HTML) content adjustments, marking a significant step forward in intelligent user interface design. The implications of this research are vast, with the potential to improve patient engagement, and satisfaction, and to address the pressing need for interfaces that adapt to diverse user behaviors and preferences.
Enhancing telehealth patient experience with emotion-sensitive large language models
This study explores the integration of Large Language Models (LLMs), specifically ChatGPT-4, to improve patient experience in telehealth. Addressing the challenge of patient anxiety during waiting periods, we implemented ChatGPT-4 for real-time emotion detection and dynamic background generation. Ex...
This study explores the integration of Large Language Models (LLMs), specifically ChatGPT-4, to improve patient experience in telehealth. Addressing the challenge of patient anxiety during waiting periods, we implemented ChatGPT-4 for real-time emotion detection and dynamic background generation. Experiments using the FACES database and qualitative feedback on generated backgrounds show that ChatGPT-4 can accurately identify emotions and create calming visual environments. These findings suggest that LLMs can significantly enhance patient engagement and satisfaction. Future developments should focus on personalized AI training and real-time adaptive systems for a more nuanced approach to patient care in telehealth.
Enhancing throughput in hyperledger fabric through endorsement policy strategy
In the realm of private permissioned blockchain platforms, increasing throughput is a pivotal objective. This paper focuses on the optimization of throughput in Hyperledger Fabric, a private permissioned leading blockchain framework tailored for enterprise applications. The paper proposes a novel ap...
In the realm of private permissioned blockchain platforms, increasing throughput is a pivotal objective. This paper focuses on the optimization of throughput in Hyperledger Fabric, a private permissioned leading blockchain framework tailored for enterprise applications. The paper proposes a novel approach to enhancing the platform’s performance by reevaluating the endorsement policy. By implementing a Less-Than-Half endorsement policy, the paper aims to streamline transaction validation processes and bridge the gap between Fabric’s throughput and the demands of large-scale industrial applications. The proposed method objects to boost transaction throughput without compromising security or reliability. The paper provides an overview of Hyperledger Fabric architecture, discusses the pre-verification mechanism, and presents the proposed method for optimizing throughput. The performance of the system is analyzed using Hyperledger Caliper and Prometheus. Simulation results show increase in the throughput of the Less-Than-Half of the endorsement policy as compared to the majority and it also demonstrates the significant reduction in the latency of the Less-Than-Half endorsement policy.
PentaPen: Combining Penalized Models to Identify Important SNPs on Whole-genome Arabidopsis thaliana Data
In the rapidly advancing field of genomics, the identification of Single Nucleotide Polymorphisms (SNPs) plays a crucial role in understanding complex phenotypic traits. This study introduces “PentaPen”, an innovative computational workflow which combines the strengths of five penalized models to ac...
In the rapidly advancing field of genomics, the identification of Single Nucleotide Polymorphisms (SNPs) plays a crucial role in understanding complex phenotypic traits. This study introduces “PentaPen”, an innovative computational workflow which combines the strengths of five penalized models to achieve improved accuracy in SNP detection. We compare the performance of PentaPen with existing models, highlighting its advantages in solving problems arising from when the number of predictors exceeds the number of samples. Beyond model comparison, we provide insights into PentaPen’s effectiveness in utilizing all SNPs as input, streamlines data pre-processing, and leverages parallel computation, enabling the workflow a considerable stride in SNP detection. Furthermore, a thorough evaluation and comparison of computational complexities signifies competitive edge of the workflow over individual penalized models. As future research directions, we propose applications of PentaPen to plant-specific characteristics and suggest further explorations to assess the robustness of its findings. In summary, this manuscript presents the genomics community with a tool that combines computational efficiency with high-precision SNP detection, making a strong contribution to the field of genomic research.
SmartCaption AI-Enhancing Web Accessibility with Context-Aware Image Descriptions Using Large Language Models
The Internet provides vast amounts of information, services, and products. However, blind individuals and those with severe vision impairments face significant challenges in navi-gating web content, especially with understanding images. This paper introduces SmartCaption AI, an innovative solution t...
The Internet provides vast amounts of information, services, and products. However, blind individuals and those with severe vision impairments face significant challenges in navi-gating web content, especially with understanding images. This paper introduces SmartCaption AI, an innovative solution that leverages Large Language Models (LLM) to generate descriptive text for images on web pages. By summarizing the content of a web page, SmartCaption AI provides relevant context for the LLM to produce accurate and meaningful image descriptions. These descriptions are seamlessly integrated into the web page's structure, allowing text- to-speech software to read them aloud to visually impaired users. SmartCaption AI offers several key contributions to web accessibility. It ensures the generated descriptions are contextually relevant, enhances the browsing experience by integrating real-time descriptions, and provides a universally accessible solution through a Chrome extension. This approach addresses the critical issue of missing or inadequate alternative text for images, thereby bridging the digital divide between sighted and visually impaired individuals. The results of our experiment demonstrated the effectiveness of SmartCaption AI, with an average score of 8.3/10, significantly outperforming state-of-art solutions: ImageToText (1.7/10) and AI-MCS (3.6/10). The source code of the tool is available on GitHub.
Tracing Economic Vibrancy: AI-Driven Analysis of Geographic Clustering in Legal Businesses
Geographic clustering of businesses holds significant importance in understanding local economic dynamics, identifying areas of commercial activity, and assisting in spatial analysis for economic development. Artificial intelligence (AI) driven analysis is employed in this paper to investigate patte...
Geographic clustering of businesses holds significant importance in understanding local economic dynamics, identifying areas of commercial activity, and assisting in spatial analysis for economic development. Artificial intelligence (AI) driven analysis is employed in this paper to investigate patterns of geographic clustering, particularly focusing on legal businesses within a given area. Data extraction techniques help preprocess business directories and classification codes to aggregate business addresses and visualize their spatial distribution. Clustering algorithms are used in conjunction with Geographic Information System (GIS) tools for data visualization and precise mapping, with respect to economic indicators. Expected outcomes include generating geographical distribution maps, comparing clustering algorithm results, and insight into urban business clustering patterns. This research considers potential external factors influencing business agglomeration and data currency. Recommendations focus on integrating AI-driven analysis with GIS tools and future research domains. Overall, this paper highlights the intersection of AI and geospatial analysis, providing stakeholders with valuable insights into the spatial distribution of economic activities within a target area.
Transforming patient experience in underserved areas with innovative voice-based healthcare solutions
This study presents a novel voice-driven online appointment system aimed at improving healthcare access in rural and indigenous communities. Utilizing existing telephone infrastructure, the system transcends the limitations of traditional chatbots by leveraging Large Language Models (LLMs) and advan...
This study presents a novel voice-driven online appointment system aimed at improving healthcare access in rural and indigenous communities. Utilizing existing telephone infrastructure, the system transcends the limitations of traditional chatbots by leveraging Large Language Models (LLMs) and advanced voice recognition technologies like Whisper. This approach enables the transformation of conventional phone calls into a streamlined digital booking process. The system’s integration with current booking processes in rural healthcare facilities and facilitates a smooth transition from voice-based to digital scheduling, providing an intuitive and efficient user experience. This innovation addresses critical healthcare accessibility challenges, notably reducing appointment booking barriers and enhancing the overall patient experience. The implementation of this technology in underserved areas demonstrates a significant advancement in patient-centered care, highlighting the role of LLMs in harmonizing traditional and digital healthcare practices. The paper provides a comprehensive overview of the system’s development, explores the challenges that may encounter, and discusses the significant potential of this approach in shaping future healthcare technologies and influencing policy decisions in healthcare accessibility.
SmartCaption AI - Enhancing Web Accessibility with Context-Aware Image Descriptions Using Large Language Models
The Internet provides vast amounts of information, services, and products. However, blind individuals and those with severe vision impairments face significant challenges in navi-gating web content, especially with understanding images. This paper introduces SmartCaption AI, an innovative solution t...
The Internet provides vast amounts of information, services, and products. However, blind individuals and those with severe vision impairments face significant challenges in navi-gating web content, especially with understanding images. This paper introduces SmartCaption AI, an innovative solution that leverages Large Language Models (LLM) to generate descriptive text for images on web pages. By summarizing the content of a web page, SmartCaption AI provides relevant context for the LLM to produce accurate and meaningful image descriptions. These descriptions are seamlessly integrated into the web page's structure, allowing text- to-speech software to read them aloud to visually impaired users. SmartCaption AI offers several key contributions to web accessibility. It ensures the generated descriptions are contextually relevant, enhances the browsing experience by integrating real-time descriptions, and provides a universally accessible solution through a Chrome extension. This approach addresses the critical issue of missing or inadequate alternative text for images, thereby bridging the digital divide between sighted and visually impaired individuals. The results of our experiment demonstrated the effectiveness of SmartCaption AI, with an average score of 8.3/10, significantly outperforming state-of-art solutions: ImageToText (1.7/10) and AI-MCS (3.6/10). The source code of the tool is available on GitHub.
Are GPTs the Answer to Small Clinics’ Digital Struggles? A Comprehensive Implementation Study
Small clinics in North America often struggle to keep pace with the digital transformation sweeping the healthcare industry due to limited financial resources and technological expertise. This digital divide has become more pronounced with the increasing reliance on digital solutions, such as online...
Small clinics in North America often struggle to keep pace with the digital transformation sweeping the healthcare industry due to limited financial resources and technological expertise. This digital divide has become more pronounced with the increasing reliance on digital solutions, such as online booking systems and telehealth services, exacerbated further by the COVID-19 pandemic. This paper evaluates whether Generative Pre-trained Transformers (GPTs), introduced by OpenAI, can effectively bridge this gap by providing a cost-effective and efficient solution for small clinics. We detail the implementation of a GPT-based online booking system tai lored to the needs of small clinics. The methodology includes a flowchart of the system’s components and descriptions, supplemented by code and scripts in the appendix. Our findings show that GPTs can significantly improve booking efficiency, reduce administrative workload, and enhance patient experience. However, we also identify drawbacks such as technical issues and the need for staff adaptation. We discuss potential issues, including error handling, privacy concerns, and appointment conflicts. The paper concludes with recommendations for small clinics on leveraging GPT technology to enhance their digital capabilities, ultimately aiming to provide more efficient and accessible healthcare services.
An Interactive Tool Enhancing Market Research Analysis through Human-AI Collaboration
The Intersection of Social Media and Public Health: A Review of Sentiment Analysis, Social Network Analysis, Trend Analysis, and Topic Modeling
ICICT 9th Congress Springer chapter
2023
Privacy, security and resilience in mobile healthcare applications
With the advent of mobile applications in health service systems, concerns such as security, privacy, usability, and resilience have been raised. We developed a system view of the concepts of security, privacy, resilience along with their relationship, and proposed a set of principles for designing ...
With the advent of mobile applications in health service systems, concerns such as security, privacy, usability, and resilience have been raised. We developed a system view of the concepts of security, privacy, resilience along with their relationship, and proposed a set of principles for designing a mobile application linking the resilience and security in privacy protection. Such study was not found in literature before. This system's view of privacy, security, and resilience has laid a foundation to develop a more effective service system. A case study is presented to illustrate how the proposed principles work in a mobile healthcare application.
A semantic model for enterprise application integration in the era of data explosion and globalisation
This paper presents a model for Enterprise Application Integration (EAI) in the modern era of data explosion and globalisation. Application here refers to software, which is in essence data system, and data refers to both information and knowledge (data serves as a vehicle for information as well as...
This paper presents a model for Enterprise Application Integration (EAI) in the modern era of data explosion and globalisation. Application here refers to software, which is in essence data system, and data refers to both information and knowledge (data serves as a vehicle for information as well as knowledge). The salient features of the model are: (1) separation of business functions from applications and enterprises, (2) three-layer architecture of the model (conceptual or semantic level, external or application level, internal or realisation level), and (3) integration of structured, semi-structured and non-structured data. To our best knowledge, the existing model or solution to EAI does not hold all the three features. A case study is presented to illustrate how the model works. The model can be used by an individual enterprise or a group of enterprises that form a network, e.g., a holistic supply chain network.
A novel scheduling method for reduction of both waiting time and travel time of patients to visit health care units in the case of mobile communication
This paper proposes a new scheduling problem for patient visits with two objectives: minimizing patient waiting time and travel time. It also presents a novel encoding method for Genetic Algorithms (GA) that is well-suited for this problem. Experiments demonstrate that the proposed encoding method r...
This paper proposes a new scheduling problem for patient visits with two objectives: minimizing patient waiting time and travel time. It also presents a novel encoding method for Genetic Algorithms (GA) that is well-suited for this problem. Experiments demonstrate that the proposed encoding method reduces optimization iterations by 17% compared to conventional methods, and the GA can decrease waiting time by up to 58.2% and travel time by up to 89.3% for specific examples. The novel scheduling problem and the encoding method are two main contributions of this work.
Adaptive user interface framework powered by a large language model for culturally sensitive virtual healthcare applications
In this research, we propose the development of anAdaptive User Interface (UI) Framework for virtual healthcareapplications, powered by a Large Language Model (LLM). Theintention is to revolutionize the way healthcare services arerendered by creating a real-time responsive system that catersto diver...
In this research, we propose the development of anAdaptive User Interface (UI) Framework for virtual healthcareapplications, powered by a Large Language Model (LLM). Theintention is to revolutionize the way healthcare services arerendered by creating a real-time responsive system that catersto diverse patient needs. Unlike conventional healthcareapplications, this framework utilizes various sensors andinteractive inputs to continuously adapt to users' feedback. Itharnesses the potential of deep learning to process this feedbackand make culturally sensitive adaptations, ensuring morepersonalized and effective care for Indigenous, Black, andPeople of Colour (IBPOC) populations. A unique aspect of thissystem is that its adaptations are not predetermined; instead, itdynamically generates changes based on the user feedbackanalyzed by the LLM. To demonstrate the efficacy of thisframework, a demo healthcare application is being developed.We expect this initiative to significantly contribute to the field ofvirtual healthcare by introducing a more inclusive, personalized,and adaptive platform, ultimately leading to improved patientcare outcomes.
Optimizing healthcare resource allocation through digital twins: a multi-objective approach for efficiency, equity, and resilience
Leveraging digital twins, this study presents an innovative approach to healthcare resourceallocation, emphasizing efficiency, equity, and resilience. Traditional methods often centralizeresources, disadvantaging rural areas. Our model, rooted in digital twin principles, addresses this byoptimizing ...
Leveraging digital twins, this study presents an innovative approach to healthcare resourceallocation, emphasizing efficiency, equity, and resilience. Traditional methods often centralizeresources, disadvantaging rural areas. Our model, rooted in digital twin principles, addresses this byoptimizing patient accessibility to services. Validated through a case study on COVID-19 test siteallocation in Saskatchewan, Canada, our approach can reduce testing disruptions by up to 92% if asite becomes inoperative. Beyond testing, the model aids in allocating critical healthcare resources,such as ICU beds and medications. While focused on healthcare, the methodology offers broaderresource allocation implications, marking a pioneering step in combining equity and resilience.
Ontology-based resilient and sustainable resource allocation and scheduling in healthcare systems: a review
The healthcare system in ageing societies such as Canada is currently facing a heavy burden due to the outbreak of COVID-19, leading to a shortage of labour and supplies. This calls for decision-makers to develop policies that can result in sustainable and resilient healthcare. To make such policies...
The healthcare system in ageing societies such as Canada is currently facing a heavy burden due to the outbreak of COVID-19, leading to a shortage of labour and supplies. This calls for decision-makers to develop policies that can result in sustainable and resilient healthcare. To make such policies, decision makers need to take a system’s view of healthcare and consider the goals of social equity and resilience, along with the efficiency of a healthcare system's operations. This review covers three aspects: (1) information modelling of a healthcare system, focusing on an ontology-based approach to information modelling, (2) concepts of system sustainability and resiliency, along with their measurements and applications in healthcare; and (3) computational approaches for resource allocation and scheduling. The review is also expected to identify the knowledge gap in each aspect and suggest future work to close the gap. At the end of this review, there is a discussion on the buzzwords of today in the field of information and knowledge systems, digital twins and big data, and their potential to add value to developing a resilient and sustainable healthcare system, which is the focus of this review.
Ontology in the Modern Computer Era
For decades, the concept of ontology has been daunting in literature. Many misconceptions exist regarding its definition. In this paper, these confusions are cleared up. A unified definition or understanding of ontology is established, which also helps to distinguish ontology, work, and work domain....
For decades, the concept of ontology has been daunting in literature. Many misconceptions exist regarding its definition. In this paper, these confusions are cleared up. A unified definition or understanding of ontology is established, which also helps to distinguish ontology, work, and work domain. Indeed, prior to the emergence of the computer, ontology is related to physics and philosophy, which does not have much confusion. Now, in the era of modern computers and informatics, ontology is connected to the computer along with the computer’s language or data (information and knowledge), which leads to confusion, as now ontology is across the human and computer. The fundamental reason for the confusion is the lack of understanding of this change, in particular the relationship between ontology and data model. This paper establishes this relationship, which makes it possible to define the so-called data model of ontology (or ontology model for short), and further to define the data model of work and the data model of the work domain. Finally, this paper examines data modelling techniques for ontology modelling, work modelling, and work domain modelling. Throughout this paper, healthcare systems are used to facilitate discussions.
Optimizing Healthcare Resource Allocation with Digital Twins: A Focus on Efficiency, Equity, and System Resilience
Healthcare resource allocation is crucial for the quality of a public healthcare system, especially resilience and equity. Conventional resource allocation methods primarily focus on efficiency, which tends to result in resources being concentrated in a geographical center. This result is especially...
Healthcare resource allocation is crucial for the quality of a public healthcare system, especially resilience and equity. Conventional resource allocation methods primarily focus on efficiency, which tends to result in resources being concentrated in a geographical center. This result is especially harmful to the situation, where healthcare facilities may experience unexpected shutdowns (due to logistic challenges and human resource shortages). Consequently, populations in rural areas face a great challenge in accessing to healthcare services. In this paper, we develop a multi-objective optimization model for representing healthcare facilities and resources along with their allocations with consideration of (1) healthcare equity, (2) resilience and (3) efficiency of services delivery. Especially, three different equity indicators for measuring accessibility to healthcare service by patients have been implemented in the model. Their impacts on the service delivery quality have been studied with a case problem, where allocation of COVID-19 test sites over Saskatchewan (Canada) is concerned. The results obtained with our model show that by considering system resilience, the allocation plan can reduce up to 92\
Uncovering the Voices Exploring BIPOC Community Views on Healthcare Policies through Social Media Data
2021
Human factors among workers in a small manufacturing enterprise: a case study
In most small manufacturing enterprise (SME), production planning and scheduling are crucial operational management tasks required. A typical make-to-order company is plagued by frequent absenteeism and abrupt resignation of skilled workers, leading to an increase in late delivery of jobs and rework...
In most small manufacturing enterprise (SME), production planning and scheduling are crucial operational management tasks required. A typical make-to-order company is plagued by frequent absenteeism and abrupt resignation of skilled workers, leading to an increase in late delivery of jobs and rework of returned jobs. The effects of some human factors, for example, job skill, job satisfaction, and job fatigue to mention a few were studied using statistical analysis. It was concluded that human factors may have significant effects on job performance in a SME. To our best knowledge, this conclusion is less known in enterprise systems in general.
2018
On rapid prototyping of assembly systems--a modular approach
This paper proposes a new product and manufacturing technology for rapid prototyping of product systems (RPPSs). It is noted that a system in this paper is defined as a physical assembly that can be decomposed into components. The rapid prototyping is achieved by a novel modular concept, that is, th...
This paper proposes a new product and manufacturing technology for rapid prototyping of product systems (RPPSs). It is noted that a system in this paper is defined as a physical assembly that can be decomposed into components. The rapid prototyping is achieved by a novel modular concept, that is, the base materials to build a component as well as an assembly is highly modularised (the interface between any two modules are identical) and building a system is simply by assembling the modules. The rapid prototyping in this paper differs significantly from the rapid prototyping in literature in that the latter builds a system layer by layer and further primarily builds a component instead of assembly (building of an assembly is actually very limited with the latter, though possible). This paper explains the RPPS approach and presents a feasibility study on the RPPS technology. It has been shown that the RPPS technology is promising.
2012
An unsupervised machine learning method for assessing quality of tandem mass spectra
In a single proteomic project, tandem mass spectrometers can produce hundreds of millions of tandem mass spectra. However, majority of tandem mass spectra are of poor quality, it wastes time to search them for peptides. Therefore, the quality assessment (before database search) is very useful in the...
In a single proteomic project, tandem mass spectrometers can produce hundreds of millions of tandem mass spectra. However, majority of tandem mass spectra are of poor quality, it wastes time to search them for peptides. Therefore, the quality assessment (before database search) is very useful in the pipeline of protein identification via tandem mass spectra, especially on the reduction of searching time and the decrease of false identifications. Most existing methods for quality assessment are supervised machine learning methods based on a number of features which describe the quality of tandem mass spectra. These methods need the training datasets with knowing the quality of all spectra, which are usually unavailable for the new datasets. This study proposes an unsupervised machine learning method for quality assessment of tandem mass spectra without any training dataset. This proposed method estimates the conditional probabilities of spectra being high quality from the quality assessments based on individual features. The probabilities are estimated through a constraint optimization problem. An efficient algorithm is developed to solve the constraint optimization problem and is proved to be convergent. Experimental results on two datasets illustrate that if we search only tandem spectra with the high quality determined by the proposed method, we can save about 56 % and 62% of database searching time while losing only a small amount of high-quality spectra. Results indicate that the proposed method has a good performance for the quality assessment of tandem mass spectra and the way we estimate the conditional probabilities is effective.
2011
Features-based deisotoping method for tandem mass spectra
For high-resolution tandem mass spectra, the determination of monoisotopic masses of fragment ions plays a key role in the subsequent peptide and protein identification. In this paper, we present a new algorithm for deisotoping the bottom-up spectra. Isotopic-cluster graphs are constructed to descri...
For high-resolution tandem mass spectra, the determination of monoisotopic masses of fragment ions plays a key role in the subsequent peptide and protein identification. In this paper, we present a new algorithm for deisotoping the bottom-up spectra. Isotopic-cluster graphs are constructed to describe the relationship between all possible isotopic clusters. Based on the relationship in isotopic-cluster graphs, each possible isotopic cluster is assessed with a score function, which is built by combining nonintensity and intensity features of fragment ions. The non-intensity features are used to prevent fragment ions with low intensity from being removed. Dynamic programming is adopted to find the highest score path with the most reliable isotopic clusters. The experimental results have shown that the average Mascot scores and F-scores of identified peptides from spectra processed by our deisotoping method are greater than those by YADA and MS-Deconv software.