32584 research outputs found
Sort by
Simple vs. Complex Human Activity Classification via Hybrid Machine Learning Models
No full textIn-depth understanding of the complexity of daily human activities is crucial for building responsive health monitoring and assistive technologies. However, limited research has focused on distinguishing activities based on their involvement level, as most existing work classifies only the type of activity performed. In this thesis, we address this gap by proposing a method to classify human activities as either simple or complex using sensor data from the Opportunity dataset. We define complex activities as those involving object interactions or multiple coordinated movements (e.g., drinking from a cup, cleaning a table), and simple activities as static or low-effort postures (e.g., standing, sitting). Based on domain-specific heuristics, we label each time window and extract both time-domain and frequency-domain features. We evaluate multiple models, including a hybrid deep learning architecture combining Convolutional Neural Networks (CNNs) and Transformers, trained on multi-sensor data across varied window lengths. To ensure robustness and generalization, we apply Leave-One-User-Out (LOUO) and Leave-One-Episode- Out (LOEO) evaluation. Our results show that the proposed approach reliably distinguishes between simple and complex activities, outperforming traditional classifiers and offering new directions for fine-grained activity recognition in real-world environments
PhotoProof: A Mobile Application for Verifying the Authenticity of Images
Get PDFThe easy access to artificial intelligence (AI) technologies, such as deepfakes and generative adversarial networks (GANs), has facilitated the creation of highly realistic artificial images, thereby undermining the authenticity of photos in today’s digital age. Misinformation and manipulation are key dangers to digital content due to this advancement. Therefore, the need for reliable methods of photo verification and authentication has become increasingly important. This report presents a decentralized iOS app that uses blockchain to ensure photo authenticity. The app leverages Ethereum smart contracts and cryptographic hashing to securely log image metadata. When a user takes a photo, the app hashes the image along with key metadata like time, location, and device details—and stores the hash on the blockchain. This creates a permanent, tamper-proof record. Users can later verify the authenticity of the image by re-uploading it to the app. The system then recalculates the hash and compares it with the one saved in the blockchain, verifying if the image has been tampered with. The app also employs Merkle Trees to support partial metadata validation, providing flexibility where some of the metadata may not be available. The solution is very cost-effective to deploy. It is inexpensive, approximately 0.00025 SEPETH per hash($0.0000326 USD), and near-instantaneous verification with no fee. These findings show that the solution is feasible and scalable, offering a cost-effective approach to maintaining photo integrity with minimal cost
Transformers in Time-Series Forecasting: Enhancing Robustness via Dynamic Attention Mechanisms
No full textTransformer architectures have emerged as powerful tools for time series forecasting, excelling at capturing complex temporal dependencies across multivariate inputs. However, these models are highly susceptible to adversarial attacks such as the Fast Gradient Sign Method (FGSM) and Basic Iterative Method (BIM), which can significantly degrade predictive performance through small, targeted perturbations. This work integrates dynamic attention mechanisms, adaptive masking modules that introduce controlled variability into attention pathways, into a transformer forecasting model to enhance robustness against such attacks. Using two distinct datasets, we compare the performance of a standard transformer and a dynamic attention-enhanced transformer under both clean and adversarial conditions. Results show that while both models perform similarly on non-attacked data, the dynamic attention model consistently maintains lower error rates under increasing adversarial intensities, demonstrating improved resilience without the need for adversarial training or additional defense layers. These findings highlight the effectiveness of dynamic architectural defenses as lightweight, model-level strategies for improving the robustness and reliability of deep learning systems in time series forecasting applications
Proxy Cap - La Lua Protector
Get PDFInspired by the object capability model and sandbox, this project, Proxy Cap, introduces a new Lua access control model that improves the language’s security without sacrificing usability. Object capability is an unconventional but powerful security model. The security model closely observes the principle of least authority. Ambient authority, the omnipresent global environment, does not exist in the object capability computation world, and no resource is accessible unless explicitly assigned. Only connectivity begets connectivity. Lua is an extensible and high-performing scripting language based on ANSI C. The language is popular in many fields but faces security challenges. Lua has a non-traditional global environment architecture that relies on a sandboxing mechanisom for its security guarantees. Sandbox is Lua’s native security tool. It offers security through isolation, but fails at being dynamic because a customized sandbox environment is static and single-use only. Proxy Cap supplements the existing sanbox by creating a new object that replaces the original object without hazardous resource access. The new object behaves like the original object but cannot perform unauthorized tasks unless explicitly given authority. This design is made possible by Lua’s C API. Real-world usecases are included in verification section to showcase the result
Multimodal Feature Fusion and Machine Learning for ADHD Detection Using Neuroimaging Data
No full textAttention Deficit Hyperactivity Disorder (ADHD) is a common neurodevelopment disorder that can significantly affect a person’s attention, impulse control, and executive function. Currently, the traditional diagnosis method often relies on clinical assessments and observations. However, these methods can be subjective and lead to inconsistencies in diagnosis between individuals. To address this challenge, neuroimaging and machine learning (ML) are promising tools for providing a more objective diagnosis of ADHD. The goal of this project is to apply a multimodal approach in which structural and functional features of specific regions of the brain are used to develop a more accurate and objective diagnostic tool for ADHD using a deep learning framework. The results of this research indicate that there are some potential but also limitations to combining the modalities together as compared to using one modality alone
Improving Contextual Retrieval for Long Documents in Q & A systems
No full textWith the vast amount of information available on the internet distributed across several lengthy documents, finding relevant information has become more important and challenging. The goal of this project is to develop advanced techniques to retrieve information from long texts in order to deliver accurate and relevant results while ensuring speed and efficiency. As part of this work, we employ techniques to address unique difficulties posed by large and complex documents. This paper presents a custom Retrieval-Augmented Generation (RAG) framework designed to improve contextual retrieval in long and multi-document settings. In this paper, we employ several techniques like summarization, semantic chunking, vectorization and knowledge graph construction to enhance query understanding and reasoning. We use the MultiHop-RAG dataset to evaluate multi-hop retrieval and question-solving scenarios where the evidence for a query is distributed across multiple documents
Strategized Reduction of Greenhouse Gas Emissions Through Predicting and Extending the Service Life of Concrete Pavements and Bridges
Get PDFStrong, durable concrete is key to resilient, long-lasting transportation infrastructure—especially in the face of climate change. This project explores innovative strategies for predicting and enhancing the service life of concrete in pavement and bridge systems, addressing the pressing need for sustainable transportation infrastructure. As concrete is pivotal to the durability and resilience of such structures, its environmental impact demands urgent attention. This project aims to reduce greenhouse gas emissions throughout their lifecycle by extending the service life of concrete pavements and bridge decks. A significant focus of the research is on the indications of chloride ion penetration in concrete, which is critical in controlling corrosion of steel reinforcing bars and keeping the infrastructure reliable. This project proposes the integration of lightweight aggregates, which have shown promise in improving the impermeability and resistance to chloride ion penetrations, which can occur through external sources such as seawater and de-icing salts. The findings provide the transportation industry with best practices and guidelines for implementing sustainable material choices while optimizing the performance of concrete structures, thereby contributing to more resilient and environmentally friendly practices that keep American infrastructure safe for moving people and goods
MAPLE: Malware Analysis through Projection of Low-dimensional Embeddings
No full textMachine learning has become a popular and powerful tool for malware analysis and detection. With the rise in popularity of natural language processing (NLP) techniques, researchers can now extract contextual embeddings from malware opcode sequences, enabling the capability to analyze hidden malware patterns and advanced code obfuscation strategies. However, unlike malware binaries, which can be directly visualized as images, these embeddings exist in high-dimensional spaces, making it difficult to observe their global patterns or spatial structures. In this paper, we propose a framework for visualizing malware embeddings in lower-dimensional space using various dimensionality reduction techniques. Our approach converts malware binaries into mnemonic opcode sequences, applies NLP models to generate embeddings, and projects these embeddings into lower-dimensional spaces for visualization. This enables us to evaluate how well different NLP techniques capture structural patterns across malware families. Experimental results show that Word2Vec outperforms BERT and GloVe in preserving both intra-family (local) and inter-family (global) structures in the reduced space. These findings are consistent with prior research highlighting Word2Vec’s effectiveness in generating meaningful malware representations. Our framework can be utilized as a visual evaluation metric that leverages low-dimensional projections to assess the quality of malware embeddings. This aids in selecting the most suitable NLP technique for capturing the structural characteristics of malware
Advancements and Applications of Eye Blink Sensor Technology: A Comprehensive Survey
No full textThis survey explores the multifaceted realm of eye blink sensor technology, emphasizing its broad applications and emerging trends. Eyeblink sensors, leveraging optical, electrode- based, and infrared methods, provide critical insights into human behavior and cognitive processes. This paper systematically exam- ines various fields where these sensors are instrumental, including human-computer interaction, cognitive load assessment, fatigue detection, neuroscience, psychology, user experience, gaming, virtual reality, and healthcare. The review delves into both hardware (e.g., Doppler, EEG, EOG) and software techniques (e.g., CAMshift, Contour Extraction), addressing the challenges and limitations inherent in eye blink detection. Emerging trends such as AI integration, data fusion, and enhanced personalization are discussed alongside ethical and privacy considerations. This comprehensive overview underscores the transformative potential of eye blink sensors, paving the way for innovative applications that enhance human interaction, cognitive understanding, and overall well-being
Improving Nurse Confidence Around Communication with Physician Partners Through the Caritas Processes
Get PDFNurses are the eyes and ears at the bedside during a patient’s hospital stay. They are responsible for communicating significant changes in a patient’s health status, mainly to their physician partners. Unfortunately, good communication between nurses and doctors is a shortcoming of the nursing profession. Seven thousand reviewed malpractice lawsuits were directly related to poor provider communication, resulting in $1.7 billion in malpractice cases and 2,000 preventable deaths. These communication errors also lead to prolonged hospital delays in care, underscoring the need for improved communication. These poor outcomes compromise patients’ lives, safety, and well-being and decrease patient satisfaction scores, ultimately reducing government reimbursement. Nurses have been described as an oppressed profession, which is mainly due to the physician hierarchy in hospitals. There is a call to action in nursing to cultivate Jean Watson’s 10 Caritas processes to help nurses improve their transpersonal relationships. The goal of this study was to provide coaching to nurses on an ICU/Tele unit on how to use Caritas daily to help improve RN/MD communication and relationships. While there was no significant statistical data to prove the intervention was effective, there were notable behavior changes in the nurses that showed a difference in the length of stay, patients’ perception of care on the unit, and positive feedback from physicians. This study suggests that if Caritas is introduced early in a nurse\u27s orientation to the organization and nursing education, nurses can have a greater impact at the bedside and have improved transpersonal relationships with their physician partners