June 27 ~ 28, 2026, Copenhagen, Denmark
Jalen Cai 1, Milin Zhu 2 ,David T. Garcia, 3 1 Diamond Bar High School, Diamond Bar, CA, 91765, 2 University of California, Los Angeles, CA, 90095
Misdiagnosis is a critical issue in global health, leading to delayed treatments, exacerbating conditions, and prolonged suffering. In the United States alone, diagnostic errors impact approximately 12 million people annually, commonly misidentifying conditions such as cardiovascular diseases, cancers, and infections. Rare diseases, affecting up to 400 million people worldwide, often receive an average of three misdiagnoses per patient before reaching an accurate diagnosis. From an economic perspective, misdiagnosis imposes a financial burden nearing $1 trillion annually in the United States for rare diseases alone, with families bearing over 60% of the costs. Systemically marginalized populations, including women and racial minorities, are up to 30% more likely to be misdiagnosed, highlighting deeply rooted societal inequities. The societal effects are compounded by clinical oversights, rushed consultations, and a lack of diagnostic inclusivity. Furthermore, environmental consequences arise from repeated diagnostic procedures and overprescription, especially in cases such as asthma, where misdiagnosis rates exceed 50%. This mismanagement leads to overuse of high-emission inhalers and improper pharmaceutical disposal, polluting aquatic ecosystems. To address these issues, we are developing an advanced interactive web application, Sympify, which integrates reputable symptom databases, including the Mayo Clinic. This application enables patients to generate comprehensive diagnostic reports based on their symptoms using a Reverse-Dictionary Algorithm. An initial experiment analyzing Sympify’s dataset found that fatigue and COVID-19 were the most reported symptoms and conditions, with symptom frequencies ranging from 1 to 160 and disease frequencies up to 214, revealing a skew toward common conditions. These findings suggest the need to balance the dataset to avoid bias in AI predictions. Future research will integrate public health data and expand Sympify’s multilingual capabilities and EHR compatibility to enhance diagnostic accuracy, reduce bias, and further minimize misdiagnosis rates.
Diagnosis, Misdiagnosis, Diagnostic Errors, Artificial Intelligence, Reverse-Dictionary Algorithm
Shako Oteka, University of Texas at Austin, USA
This paper proposes a conceptual framework for understanding modern artificial intelligence systems through African ancestral cosmological models. Rooted in the author’s Batetela cultural heritage from the Democratic Republic of Congo and a first-person AI/ML learning journey, the paper maps five AI architectural functions — data pipelines, APIs, automation, security, and state management — onto five African cosmological concepts: Mizimu, Simbi, Nkondi, Kaya, and Kalunga. Treating these frameworks as interpretive models rather than spiritual claims, the paper contributes to discourse on culturally inclusive technology pedagogy, indigenous knowledge systems, and human-centered AI design. The central thesis is that the organizational logic underlying AI systems mirrors long-standing human frameworks for understanding memory, flow, action, protection, and transformation.
Artificial Intelligence, African Cosmology, Indigenous Knowledge Systems, Human-Centered Design, Systems Thinking, Technology Pedagogy, Batetela, Kongo Philosophy
Sarra Ayouni, Raneem Alkhonain, Lamya Alosaimi, Abeer Alghanem, and Sadeem Ababtain, Princess Nourah bint Abdulrahman University, Saudi Arabia
Dyslexia is one of the most common learning disabilities. It affects Children worldwide , and impacts their reading fluency, spelling accuracy, and comprehension. Arabic-speaking children with dyslexia face significant challenges due to the linguistic complexity of the Arabic language, including diacritics, connected letter forms, and right-to-left script orientation. Despite increasing global attention toward assistive educational technologies, Arabic-focused digital solutions remain limited. In this paper, we propose a gamified educational mobile application designed to support Arabic-speaking children with dyslexia by providing structured practice in reading and spelling through engaging and interactive games. The application integrates evidence-based dyslexia strategies, speech-to-text (ASR) technology for pronunciation evaluation, and progress-tracking tools accessible to parents and experts. It also enables communication between parents and specialists, in addition to providing supportive guidance through an AI-based assistant. Through questionnaires and interviews with experts and parents, we have identified key user needs and highlighted the importance of gamification in enhancing motivation and improving reading and spelling skills. Our proposed system aims to bridge the gap in Arabic dyslexia resources by offering an inclusive, enjoyable, and effective digital learning experience.
Dyslexia, Arabic language, Gamification, Assistive technology, educational games, Speech recognition, reading skills, spelling skills, child engagement, learning difficulties.
Daad M. Alhassan, Hala M. Alsuabey, Rana A. Almashari, Rand A. Altareefi, Layan A. AlRushaid, and Fahima Hajjej, Princess Nourah bint Abdulrahman University (PNU), Riyadh, Saudi Arabia
Emergency department (ED) overcrowding is a persistent operational challenge that degrades patient care quality, lengthens waiting times, and strains hospital resources. Accurate forecasting of patient arrivals enables proactive staffing, bed allocation, and capacity planning. This paper presents a comparative study of machine learning and deep learning models for forecasting daily ED visit volume, developed within a modified CRISP-DM framework. Using de-identified records from the MIMIC-IV-ED (v2.2) database enriched with historical weather observations and calendar features, we engineer temporal predictors and evaluate five models: a Long Short-Term Memory (LSTM) network, a Random Forest Regressor (RFR), Extreme Gradient Boosting (XGBoost), Neural Basis Expansion Analysis for Time Series (N-BEATS), and the Temporal Fusion Transformer (TFT). Models are assessed with Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and Mean Absolute Percentage Error (MAPE) on a chronologically held-out test set. The TFT substantially outperformed all baselines, achieving an MAE of 4.46, an RMSE of 5.02, and a MAPE of 41.5%, compared with MAE values of 12.8-14.2 for the remaining models. To promote transparency, SHapley Additive exPlanations (SHAP), Partial Dependence Plots (PDP), and the TFT attention weights are used to interpret model behaviour. The best model is embedded in an interactive forecasting dashboard, "Marsad," that produces daily and short-horizon hourly forecasts to support proactive ED operations in alignment with Saudi Vision 2030.
Emergency department forecasting, temporal fusion transformer, machine learning, deep learning, explainable AI, time-series prediction, healthcare operations.
Jose Raul Leggs Peralta , Mexico
We report experimental observation of a kink structure in qubit T2 coherence decay at a characteristic time tk = alpha x T2, where alpha = 7.2973525693 x 10-3 is the fine-structure constant. CPMG-1 measurements on IBM Kingston Q33 (real superconducting hardware) yield friction factor f = 1.196 with confirmed directionality T2 a > T2 b , and empirical k-parameter k emp = 0.94 consistent with IRT prediction k = 1. We introduce the IRT Coherence Scheduler, a hardware-agnostic tool that applies revival pulses at intervals tk: T1,eff = 484.5 us (1.50x), T2,eff = 619.5 us (1.196x), and 29.5x improvement over the gate-limited baseline. IRT simulation at T2 = 500 ms yields DeltaAIC = 15.57 and kemp = 0.988. Platforms with T2 ~ 1-10 s exhibit tk in the 7-73 ms range with projected 66x coherence enhancement, directly benefiting fault-tolerant quantum computation and quantum chemistry simulation.
quantum decoherence, fine-structure constant, CPMG, coherence scheduler, trapped-ion quantum
Katleho Moloi, Centre for Augmented Intelligence and Data Science, University of South Africa
Cloud computing infrastructures are increasingly being deployed in regions where power supply is intermittent and unreliable, posing significant challenges to service continuity and efficient resource utilization. Conventional cloud scheduling strategies typically assume stable energy availability and therefore perform poorly under fluctuating power conditions. This paper proposes an Energy-Adaptive Predictive Scheduling (EAPS) framework designed to enable resilient cloud operation in intermittently powered environments. The proposed architecture integrates short-term energy forecasting with adaptive workload orchestration and energy-aware node management, allowing computing workloads to be dynamically aligned with predicted energy availability. The scheduling problem is formulated as a multi objective optimization task that maximizes service availability and workload completion while minimizing energy consumption and migration overhead. To solve this problem efficiently, a lightweight predictive scheduling algorithm is developed to perform proactive workload consolidation and migration under energy constraints. Experimental evaluation demonstrates that the pro-posed approach significantly improves service availability, workload completion rate, and node utilization compared to conventional scheduling strategies. The results highlight the potential of predictive energy-aware orchestration for enabling resilient and sustainable cloud computing infrastructures in power-constrained environments.
Energy-adaptive cloud computing, Intermittent power systems, Energy-aware scheduling, Predictive resource management, Sustainable cloud computing.
Xin Wen 1, Garret Washburn, 1 USA, 2 California Baptist University, USA
Prescription stimulant misuse remains common among U.S. college students, with measurable harms to sleep, cardiovascular health, and downstream academic performance, yet most existing prevention tools are static brochures or single-session workshops that under-engage digital-native young adults. This paper presents MindBalance, a Flutter mobile application that integrates four components — a Daily Reflection check-in, a curated ten-lesson Brain Science library, a personalized 7-day Challenge game powered by OpenAI’s gpt-4o-mini, and an editable user Profile — through Firebase Authentication and Cloud Firestore. The Challenge prompt is conditioned on the user’s age and most recent Daily Reflection, so scenarios match current vulnerability (e.g., a sleep-deprivation dilemma after a fourhour night). Internal validation showed personalized prompts produced substantially higher contextual relevance than non-personalized or hardcoded baselines, and simulated playthroughs confirmed the game’s stat-delta mechanics encode the intended educational message. The system is a reproducible template for adaptive, AI-assisted health-behavior interventions.
Prescription stimulant misuse, Mobile health (mHealth), Large language models, Gamification
Annika Weibe, Richard George, Stephan Henker, and Christian Mayr, Department of Electrical Engineering and Information Technology,Technical University of Dresden (TUD), Germany
This paper presents a real-time signal processing architecture for processing electrode recordings from neuronal tissue. In combination with an IBEX RISC-V CPU, two dedicated hardware accelerators are implemented for state-of-the-art on-chip neural signal processing tasks: a Multiply-Accumulate (MAC) unit and a Convolution engine (CONV), both highly configurable and optimized for signed fixed-point arithmetic. The MAC unit significantly enhances on-chip processing efficiency by offloading computationally intensive operations from the CPU. To enable fully autonomous, on-chip spike sorting, a lightweight Principal Component Analysis (PCA) routine has been implemented directly on the CPU. This architecture enables on-chip feature extraction and training. This on-chip training ca pability eliminates the need for offline preprocessing and enables real-time adaptation to changes in the recording due to electrode drift, artifacts, etc. We demonstrate the flexibility and power of the accelerators by implementing multiresolution spectral analysis and com petitive spike sorting, leveraging the PCA-based feature extraction pipeline. Employing the MACunit for inference reduces the energy consumption from 1.43µJ/spike to 1.09µJ/spike, representing a 23.8% improvement in energy efficiency compared to a pure CPU implemen tation. The Convolution engine supports band-selective filtering and multiresolution Discrete Wavelet Transform (DWT) decomposition, facilitating on-chip time-frequency analysis and enabling up to 75% reduction in off-chip data bandwidth. The chosen processing tasks under line the utility of convolution and filter accelerators in particular, when it comes to designing a flexible and hardware-efficient computing platform for neuronal signal analysis. The pre sented System-on-Chip implemented in 22nm fully-depleted silicon-on-insulator (FDSOI) technology, facilitating ultra-low-voltage operation at 0.55V.
Neural Signal Processing, neural prosthetics, System-on-Chip, Spike Sorting, Multiply-Accumulate unit, Finite Impulse Response filter, Principal Component Analysis, Convolution, Multiresolution Discrete Wavelet Transform.
Yufei Zhang 1, Yu Sun 2, 1 Troy High School, 2200 Dorothy Ln, Fullerton, CA 92831, 2 California State Polytechnic University, Pomona, CA 91768
Global waste generation continues to accelerate while recycling infrastructure struggles with inefficiency, contamination, and labor challenges that limit material recovery rates below 35% in most regions. This research presents an intelligent automated recycling system integrating YOLO-based computer vision, precision robotic manipulation, and mobile application interfaces to address multiple dimensions of the recycling challenge simultaneously [1]. The system employs a dual-model detection architecture operating at 30 FPS to identify bottles, cans, and cups, triggering material-specific robotic sorting sequences through a servo-controlled mechanical arm. Experimental validation demonstrated 84.4% detection accuracy across diverse environmental conditions and operational throughput of 90.6 items per hour—exceeding manual sorting performance by 51-127%—while maintaining total error rates of 6.3%. The integrated Flutter mobile application provides users with recycling value transparency and environmental impact visualization, promoting behavioral engagement [2]. Comparative analysis against recent methodologies reveals that the unified architecture approach achieves superior throughput and reliability compared to separate-module systems while complementing logistical optimization platforms. This comprehensive solution demonstrates that combining computer vision automation with user education creates a scalable pathway toward sustainable circular economy practices.
Automated Recycling, Computer Vision, Robotic Manipulation, Sustainable Systems
Jiaxuan Li 1, Jonathan Sahagun 2, 1 The Loomis Chaffee School, 4 Batchelder Road, Windsor, CT 06095 , 2 California State University, Los Angeles, 5151 State University Dr, Los Angeles, CA 90032
Food insecurity is increasingly understood to involve not only access to food, but the nutritional literacy required to make healthful decisions, a literacy that is unequally distributed among low-income households, older adults, and people managing diabetes or hypertension. Existing nutrition apps assume time, connectivity, and the willingness to type every meal, which limits their reach. NutriLens is a wearable smart-glasses system that delivers real-time nutritional guidance hands-free by combining a Raspberry Pi camera, a Flutter iOS application, a custom Bluetooth Low Energy image-transfer service, Google Gemini vision-language inference, and a Vufine heads-up display. Challenges in BLE framing, JSON robustness, and per-user data isolation were resolved through chunked typed packets, schema-constrained prompts, and UID-scoped Firebase paths [5]. Across two experiments the system achieved a mean 5.5% calorie estimation error and a 13-second image transfer latency at default settings. NutriLens reframes nutrition tracking from a typing task into an ambient one.
Gemini, Vision-language model, Bluetooth Low Energy, Flutter, Raspberry Pi
Junxi Pan 1, Garret Washburn 2, 1 Corona Del Mar High School, 2101 Eastbluff Dr, Newport Beach, CA 92660 , 2 California Baptist University, 8432 Magnolia Ave, Riverside, CA 92504
Rowing technique determines both performance and injury risk, yet most athletes outside elite programs lack access to the marker-based motion capture or full-time coaching that would let them quantify their form. This work presents StrokeSense, a mobile application that analyses rowing technique from ordinary smartphone video. A Flutter front end uploads a clip to a Flask backend, where MediaPipe Pose extracts 33 body landmarks per frame, a biomechanical engine computes layback, leg, shin, elbow, and forward-layback angles, and a GPT-4 assistant turns the per-frame angle history into a structured coaching report delivered through Firebase. Experiments show landmark detection accuracy above 88% in every tested condition and a mean absolute joint angle error of 2.8°, well below the 5° threshold that coaches use informally. By turning a single phone recording into quantitative, narrated feedback, StrokeSense makes rigorous rowing analysis broadly accessible.
Rowing biomechanics, MediaPipe, OpenCV, Flutter, Firebase
Xinhao Gao 1, Samuel Silverberg 2 , 1 Portola high school, 1001 Cadence, Irvine, CA 92618, 2 California State Polytechnic University, Pomona, CA 91768
Retail investors lack a low-friction tool that combines live cross-sector price comparison, personal buy-price anchoring, and transparent decision support in a single browser-native view. The proposed Live Multi-Stock Tracker addresses this gap through a static HTML and JavaScript application hosted on Firebase Hosting, integrating a Chart.js animated line chart, a per-stock table with user-saved buy prices and percent-since-buy indicators, and an AI advisor that posts recent ticks to the Gemini API and renders the model’s recommended ticker with a confidence score and rationale. Engineering challenges around update jank, persistence of saved references, and API rate limits were addressed through bounded data windows, planned local-storage persistence, and prompt debouncing. Two experiments evaluated advisor hit rate and percent-since-buy correctness, yielding a 20 percent advisor hit rate near the 16.7 percent random baseline and 100 percent arithmetic agreement. The system provides retail users with an accessible, transparent, real-time monitoring dashboard at near-zero deployment cost.
Real-Time Stock Tracking, Financial Dashboard, Large Language Model, Gemini API, Chart.js, Firebase Hosting, Browser-Based Visualization, Personal Portfolio Monitoring
Alvin Zhu 1, Rodrigo Onate 2 , 1 Fairmont Private School, 26333 Oso Rd, San Juan Capistrano, CA 92675, 2 California State University, Fullerton, 800 N State College Blvd, Fullerton, CA 92831
The increasing reliance on digital technology has created a growing demand for accessible and reliable technical support, particularly among individuals with limited digital skills. This paper addresses this issue by proposing HelpHub, a mobile application designed to facilitate peer-to-peer technical assistance through real-time interaction and artificial intelligence. The system integrates three primary components: a user interface built with Flutter, a cloud-based backend using Firebase, and an AI Help feature powered by natural language processing [8]. Several challenges were identified during development, including ensuring secure authentication, maintaining reliable data storage, and improving the accuracy of AI-generated responses. These challenges were addressed through structured system design, validation mechanisms, and prompt optimization strategies. Experimental evaluations were conducted to assess AI performance and database reliability, revealing generally strong results with some areas for improvement, particularly in handling ambiguous inputs and unstable network conditions [9]. Overall, HelpHub demonstrates that combining community-driven support with AI assistance creates a scalable, efficient, and cost-effective solution. By reducing barriers to technical help and supporting multiple languages, the system has the potential to significantly improve digital accessibility and user confidence in technology.
Help, Volunteer, Solving problems, Technology
Abhishek Bhardwaj, Max Quinn, and Apeksha Agnihotri ,Amazon Web Services,New York, NY, USA / San Francisco, CA, USA
This paper presents SALT (Sustained Anomaly Labeling in Time Series), a novel approach combining autoencoder neural networks with gradient-based filtering for detecting sustained anomalies in time series data. While traditional anomaly detection methods often struggle with distinguishing between transient spikes and meaningful sustained anomalies, SALT addresses this limitation by leveraging autoencoders’ ability to learn normal data patterns and incorporating gradient filtering to identify persistent deviations. The method was evaluated on both curated data and real-world datasets from the Numenta Anomaly Benchmark (NAB). Results demonstrate significant improvements in precision while maintaining high recall - achieving F1 scores of 0.91 and 0.86 on curated and real-world data respectively, compared to 0.72 and 0.48 using autoencoders alone. SALT successfully reduces false positives from transient anomalies while accurately detecting sustained anomalous patterns. The framework shows promise for applications in network security, industrial systems monitoring, and other domains requiring reliable detection of persistent anomalies in streaming data.
Sustained Anomaly, Gradient Filtering, Deep Learning, Data Labeling, Time Series, Autoencoder
Kwan Chak Stephen Sun 1, Austin Amakye Ansah 2 , 1 Pacific Ridge School, 6269 El Fuerte St, Carlsbad, CA 92009, 2 The University of Texas at Arlington, 701 S Nedderman Dr, Arlington, TX 76019
Dialect is a cross-platform mobile application designed to bridge the gap between language learning and cultural discovery. Built with Flutter and Firebase, the application integrates three core components: a Word Studio for vocabulary acquisition with cultural context, a Learning Lab featuring AI-powered conversation practice with realtime feedback, and a social feed for community engagement and cultural sharing [1]. The application addresses key challenges in mobile language learning including notification management, real-time data synchronization, and AI conversation quality. By combining structured learning with social interaction and AI assistance, Dialect provides a comprehensive platform for developing both linguistic proficiency and cultural understanding [2].
Cultural discovery, Language learning, Mobile application, Flutter, Firebase