Figure skating is a physically demanding and expensive sport, and beginners often lack enough accessible support to practice safely and improve efficiently. Skate Sensor addresses this problem through a wearable sensing system paired with a mobile application that helps users analyze skating techniques and access educational resources [1]. The project combines two shin-mounted sensor pods, a Flutter mobile app, and machine learning models developed with Python and PyTorch [2]. Together, these components collect motion data, synchronize left and right leg activity, classify techniques, and evaluate whether an attempt matches successful motion patterns. Several implementation challenges had to be considered, including sensor synchronization, limited training data, and memory use within the app’s media features. These issues were addressed through timeline alignment, expanded data collection strategies, and more efficient loading of visual content. Experimental design focused on model accuracy and the impact of informational screens. Overall, Skate Sensor demonstrates strong potential as a practical tool for safer, more informed, and more independent beginner training.
Optical camera communication (OCC) is considered as a key enabler of optical wireless communication technology. In OCC, light-emitting diodes (LEDs) serve as the transmitter and rolling shutter (RS) cameras as the receiver for high-speed communication. However, the received luminance from the LED is critically important for reliable data retrieval in OCC, which faces challenges due to the inherent nature of data collection. Smartphones, typically equipped with RS cameras, represent one of the most promising platforms for the commercial deployment of OCC. To ensure system reliability, various methods have been proposed to address the diversity in pixel illumination values captured by smartphone cameras. Furthermore, AI-based approaches that make RS cameras compatible with low-speed mobile scenarios and enhance overall system performance also introduce additional system complexity. In this paper, we provide a systematic review of the state-of-the-art methods for data retrieval in smartphone camera-based OCC. In particular, we provide specific challenges due to important factors, such as communication distance variation and the blooming effect. Furthermore, we discuss recent advancements, especially promising AI applications in OCC. Finally, we outline open research directions on smartphone camera-based OCC.
Manual inspection of photovoltaic systems is expensive, hazardous, and prone to inconsistency. This paper presents SunScout, a mobile application for offline solar-panel image management and on-device fault classification. The current mobile release organizes drone, gallery, and camera captures into reusable datasets and analyzes each stored asset with a fine-tuned EfficientNetB0 classifier deployed through ONNX Runtime. On the cleaned 1,575-image dataset, a frozen MobileNetV2 baseline reached 90.79% validation accuracy, while the proposed EfficientNetB0 model achieved 94.29%, a 3.50 percentage point improvement. Together, these results show that accurate solar fault analysis can be delivered on a consumer smartphone without requiring a persistent network connection.
Neurological and psychiatric conditions including Alzheimer's disease, PTSD, depression, and schizophrenia affect hundreds of millions of people globally, yet existing pharmaceutical treatments are expensive, inconsistent, and out of reach for many. This paper proposes frequency-based music therapy, delivered through an AI-powered mobile application called Querey, as a clinically grounded and non-invasive alternative. Querey is built around three components: a state-based discovery survey, an adaptive AI coach, and a mood stimulation engine rooted in brainwave entrainment and BPM science [9]. Challenges included music licensing constraints, navigating App Store deployment as a first-time developer, and maintaining data accuracy across the personalization pipeline. Experiments showed a mean satisfaction score of 7.5 out of 10 for BPM accuracy, with calming prescriptions outperforming energizing ones, and a twelve-week clinical trial comparing Querey against live therapy and passive listening across diagnosed populations [1]. When technology is designed around how the brain actually works, the results speak for themselves.
This project develops a Unity-based simulation platform that helps students and beginner designers analyze vehicle aerodynamics without expensive wind tunnel testing [6]. Air resistance plays an important role in vehicle performance, but traditional aerodynamic testing methods require costly equipment and complex professional software. To address this problem, the project introduces an interactive simulation environment where users can upload 3D car models, select or upload maps, and simulate airflow conditions [7]. The system is built around three main components: Properties, CFD simulation, and Map. These components work together to control inputs, generate aerodynamic visualization, and manage user interaction. The CFD system uses a simplified real-time approach based on dynamic pressure and precomputed surface responses to provide stable and interactive feedback. During development, challenges such as path design, model scaling, and collision detection were addressed through physics adjustments and automated systems. Experimental evaluation compared the system against ANSYS Fluent, showing an average error of 5.5%, indicating that the platform provides reasonably accurate results while maintaining high performance and accessibility.
Parkinson's disease creates a long-term rehabilitation problem because patients often need frequent movement therapy, yet access, adherence, and motivation remain difficult to sustain. MirrorMove PD is a prototype therapeutic dance system designed to address that problem through a Flutter mobile app, a Python desktop companion, and a local HTTP control bridge. The mobile app manages authentication, goals, progress, and song selection, while the desktop component displays guided choreography and session playback. Experimental scripts using MediaPipe and reference landmark extraction support future movement analysis and scoring. The project must address three major challenges: reliable pose comparison, dependable phoneto-desktop communication, and meaningful workout metrics. Two preliminary approximate experiments suggest that trust in scoring depends on keeping feedback close to user expectations and that repeated use may improve short-term confidence. Overall, the prototype is promising because it combines evidence-informed dance rehabilitation with a practical homeuse delivery model that is accessible, structured, and expandable.
California agriculture faces the combined pressures of severe drought, high crop-waste rates, and unaffordable commercial precision-agriculture platforms, which together disproportionately affect small and mid-sized growers. This paper proposes an integrated planthealth monitoring platform that combines a Raspberry-Pi field node for image capture and environmental sensing, a multimodal vision model for species identification and health assessment, and a Flutter mobile client that presents results to the grower through a simple dashboard. The client implements a layered fallback between the live Pi, an on-disk cache, and a bundled sample dataset so that it remains functional under intermittent connectivity, and it caches plant images transparently to accelerate repeated views. Two experiments evaluated the system: species identification reached 87.5 percent accuracy across four visually similar species, and time-to-first-paint ranged from 0.38 seconds on cached data to 1.34 seconds under degraded networks. The platform demonstrates that practical precision agriculture is achievable at consumer-hardware scale.
Students increasingly struggle to engage effectively with study materials, with only 34% reporting active learning engagement and 65% experiencing academic anxiety. AIvy is an AIpowered web application that addresses this dual challenge by transforming uploaded study materials into personalized, interactive learning experiences while tracking student wellness. The system leverages OpenAI’s GPT-4o model to extract text from diverse document formats using vision capabilities, analyze content to generate targeted summaries and five-question quizzes aligned with user-specified learning objectives, and recommend educational YouTube videos [9]. A parallel wellness journaling system tracks daily mood, study preferences, and focus patterns, generating personalized study recommendations. Built on a serverless architecture with Vercel Python functions and Firebase for authentication and data persistence, AIvy ensures API key security while maintaining responsive performance [10]. Experimental evaluation demonstrated 88.2% mean quiz quality and 88.6% text extraction accuracy across document types, confirming the system’s viability as a comprehensive educational companion.
Simulating public opinion evolution is a core focus of computational social science. Traditional agent-based models rely on predefined heuristic rules, failing to capture the semantic features and cognitive processes of human natural language interactions. While large language models offer new approaches for artificial society construction, existing frameworks have limitations in scalability and memory management. Taking the Fukushima nuclear wastewater discharge event as the background, this study uses an open-source multi-agent social simulation framework, designing four progressive intervention scenarios to analyze agents' cognitive synergy and public opinion trajectories. Results show the framework mitigates role drift and premature consensus, reproduces the public opinion evolution trajectory, providing empirical insights for policy testing and LLM-driven social computing.
Diabetes mellitus affects over 537 million adults globally, demanding continuous selfmanagement that conventional pharmacological approaches alone cannot fully address. Music therapy has emerged as a promising complementary intervention, with clinical research demonstrating that slow-tempo music can reduce blood glucose by 15-30 mg/dL through parasympathetic activation and cortisol reduction. BeatSugar is a cross-platform mobile application that integrates real-time blood sugar and heart rate monitoring with personalized, evidence-based music therapy recommendations. The system employs a context-aware algorithm that maps blood glucose levels, measurement timing, and diabetic status to clinically appropriate music tempos, incorporating Traditional Chinese Medicine Five Element tonal sequences alongside AI-generated therapeutic compositions. A personalized effectiveness scoring engine learns from individual listening sessions, adapting recommendations based on measurable health outcomes. Experimental evaluation demonstrates 94.2% recommendation accuracy and algorithm convergence within 8-12 sessions. BeatSugar offers a scientifically grounded, scalable approach to complementary diabetes management through accessible digital music therapy.
Science museums struggle to provide personalized, age-appropriate experiences for diverse family audiences, leading to suboptimal learning outcomes and visitor engagement. This paper presents an AI-powered mobile companion application for Discovery Cube Orange County that addresses these challenges through intelligent personalization and interactive guidance. Built with Flutter and OpenAI’s GPT-4, the system integrates three core components: a profile management system storing child demographics, an AI recommendation engine generating personalized exhibit suggestions, and an interactive tour system combining QR code scanning with conversational AI assistance[8]. Implementation challenges included managing API response latency, ensuring age-appropriate content accuracy, and handling variable network conditions. Experimental evaluation across 60 age-appropriateness trials achieved a mean rating of 4.45/5.0, with strongest performance for ages 8-10 [1]. Network performance testing revealed bandwidth as the critical factor, with response times ranging from 1.8 to 5.4 seconds. The system demonstrates that AI-driven personalization makes museum experiences more accessible and educationally effective than traditional AR-based or static content approaches, offering a scalable model for informal STEM education enhancement.
General aviation pilots frequently face fatal accidents due to cognitive overload when manually interpreting complex weather and terrain data mid-flight. To solve this, we developed SkyAware, an active, intelligent flight companion. Built with Flutter, the application integrates the Gemini API, Open Maps Terrain, and AviationWeather data to provide real-time 3D hazard monitoring and conversational safety briefings [1]. Core challenges included accurately synchronizing asynchronous APIs, ensuring AI hazard analysis reliability, and visualizing dense spatial data. We mitigated synchronization latency by implementing predictive forward vectoring. Experimentation using historical NTSB events and live MSFS 2024 telemetry yielded our most important results: an 86% mean AI hazard detection accuracy and a highly responsive 1,340ms average warning latency [2]. The results indicate that SkyAware effectively translates raw metrics into actionable insights, improving pilot decision-making. By proactively preventing alert fatigue, SkyAware empowers safer mid-flight decision-making and save lives in the cockpit.
By Zadeh’s original formulation, likelihood distributions can be seen as a unique kind of fuzzy set whose logical operations have meaningful probabilistic interpretations. In this work, we develop a variant of this fuzzy set in which an arbitrary number of fuzzy logic operations may be applied without increasing the space and time required for membership evaluation. By using a band-limited Fourier series approximation for a truncated Gaussian kernel, we also demonstrate a significant reduction in time and space requirements for mixture model evaluation. Probabilistic and fuzzy set interpretations, as well as benchmarks against Standard Additive Models (SAMs) are provided along with an analysis of complexity and scaling properties
Entity resolution (ER) typically relies on pairwise similarity comparisons between records, which limits its ability to capture indirect relationships present in demographic occupancy data. An important indirect pattern arises from household movement, where multiple individuals relocate together across addresses, but detecting such patterns is difficult due to mixed-format records, noise, duplication, and the absence of stable identifiers. This paper proposes an AI-enhanced framework for detecting indirect entity links associated with household movement in unstandardized name-address data. The approach integrates prompt-based large language model (LLM) named entity recognition for extracting personal names and addresses without extensive preprocessing, semantic text embeddings for robust similarity computation, and graph-based reasoning to infer group-level movement patterns. Experimental evaluation on SPX benchmark datasets (S8-S12) generated using the Synthetic Occupancy Generator demonstrates that incorporating indirect household movement evidence improves recall by 8-15
Large language models (LLMs) sometimes generate convincingly written but factually incorrect content, commonly referred to as hallucinations. We introduce an algorithm, Magic Numbers (MN), which is compatible with any LLM or LLM API that outputs token probabilities (at the time of writing, only certain GPT-models). It uses combinations of perplexity and logit entropy to detect and fix hallucinations in real time. We show that MN can detect hallucinations in GPT–4o with only a negligible addition to the computational cost. Our experiments use two frameworks for hallucination measurement. HaluBench (which focusses on question-answering) and HHEM (which targets summarization tasks). The results indicate that Magic Numbers can improve the veracity of GPT–4o in various question-answer tasks and also improve the factual consistency in text summarisation.
In this paper, we propose a novel method for rotation invariant pattern recognition. We perform adaptive denoising to the input pattern images. If the noise level is above a threshold, we perform block matching and 3D filtering (BM3D) to reduce noise from the noisy pattern images. We do not conduct denoising otherwise. We extract ridgelet-Fourier features from the denoised pattern and classify the unknown pattern to one of the known classes with the nearest neighbor classifier. Experiments demonstrate that our new method achieves perfect classification rate (100
Texture classification is a critical task with applications spanning various domains, from facial recognition to cancer detection in medical images. In traditional approaches, the application’s success heavily depends on the feature extraction and classification stages. Over the years, numerous feature extraction methods have been proposed, with non-handcrafted approaches consistently outperforming handcrafted ones. This paper investigates static and dynamic selection techniques of classifiers trained on features extracted from non-handcrafted architectures. The experiments were conducted on two challenging benchmarks widely used for texture classification evaluation: the FMD dataset and the Describable Texture Dataset (DTD). We first evaluated individual features and found that Visual Transformers (ViT) performed exceptionally well compared to other architectures. However, the significantly higher Oracle accuracy for the ensemble of classifiers suggests room for improvement in investigations concerning classifier combination and selection techniques. We observed enhanced performance when combining the top-performing individual classifiers by applying static combination methods such as sum, product, and max rules. Dynamic classifier selection techniques have not yielded improvements in the rates. The best performance was achieved using a static combination of classifiers through sum and product rules. In the FMD database, the F1-score was 93.2
Classifying gravitational wave signals is an essential task in analyzing data from space collected by advanced tools such as an interferometer. In this paper, we present a new architecture of a convolutional neural network that classifies gravitational wave spectrograms into a selected class. For this purpose, a novel attention module based on the fuzzy controller architecture was proposed. The mechanism is based on the generation of matrices: query Q, keys K, and values V, where the first two are fuzzified by a Gaussian function and subjected to fuzzy inference. The inference results are sharpened and multiplied by values in V. This solution allows the use of the idea of a fuzzy controller to analyze features in neural networks. The model was tested and analyzed in terms of different evaluation metrics that show that this model can reach higher results than the state-of-the-art.
In the era of smart cities and AIoT infrastructure, deploying efficient machine learning models on resource-constrained edge devices has become critical for urban utility management. This paper evaluates the effectiveness of various data augmentation techniques in enhancing the performance of machine learning models on these devices despite their limited computational resources. Our study utilizes three datasets of digit images: one from Kaggle, one from SCUT, and a proprietary dataset. We tested ranges of parameters for data augmentation, including noise, brightness, contrast, and geometric transformations, to assess their impact on model accuracy. The findings indicate that while augmentation generally improves model performance, an optimal range exists beyond which accuracy may decline due to overfitting. This paper describes this standardized approach to parameter testing that contributes to developing more efficient and accurate edge-based machine learning applications.
People with the autism spectrum have special needs when it comes to using computers. Therefore designers of graphical interfaces dedicated to such people should bear in mind the limitations faced by them. In this paper, the system, which supports designers by automatically evaluating the accessibility of graphical interfaces, is proposed. Image processing methods are used to extract features, on which the introduced fit scale allowing for evaluation of interface adaptation to special preferences is based. The presented approach facilitates the creation of interfaces intelligible for autistic people and raises awareness about their needs. The fit scale values were calculated for several game interfaces and the obtained results have been compared with a visual analysis performed by an expert working with autistic people.
