keyboard_arrow_up
Accepted Papers
Simulation-Based Evaluation of a Cross-Jurisdictional Compliance Index for Digital Medical Devices

Kwang-Man Ko, Department of Computer Engineering, Sangji University,83 Sangjidae-gil, Wonju-si, Gangwon-do, Republic of Korea

ABSTRACT

This study proposes a traceable method for converting multi-jurisdictional cybersecurity and software-quality requirements into two quantitative compliance indicators for digital medical devices and evaluates its methodological validity through simulation. Requirements from the Korean MFDS, the U.S.FDA, the EU MDR and MDCG guidance, IMDRF guidance, and relevant IEC, ISO, and AAMI standards are consolidated into traceable verification items. A compensatory Weighted-Sum indicator summarizes overall development progress, whereas a Jurisdiction-Bottleneck indicator identifies the lowest jurisdictionspecific level of compliance. A simulation-based evaluation using a ten-item insulin-pump scenario examines the calculation procedure and decision-making behavior of the proposed indicators. The simulation yields Weighted-Sum and Jurisdiction-Bottleneck scores of 69.0% and 60.0%, respectively. Sensitivity analysis using alternative enforceability encodings confirms the stability of the identified bottleneck, while counterfactual analysis distinguishes item-level improvement effects from denominator effects. These results demonstrate the computational consistency, interpretability, and complementary decision behavior of the proposed indicators. However, the simulation does not establish industrial effectiveness, inter-rater agreement, or the ability to predict actual regulatory approval outcomes.

KEYWORDS

Digital Medical Devices, Medical Software, Standard Compliance Evaluation, Quantitative Evaluation Index


Hearassist: Design and Evaluation of an AI-Powered Assistive System for Real-Time Speech Transcription and Environmental Sound Detection

Haowei Sun1, Bobby Nguyen2, 1TVT Community Day School, 5 Federation Way, Irvine, CA 92603, 2University of California, Riverside, 900 University Ave, Riverside, CA 92521

ABSTRACT

Deaf and hard-of-hearing individuals can face barriers when accessing spoken information and important environmental sounds. hearassist addresses this problem through an AI-powered mobile system combining live transcription, conversation summaries, environmental sound detection, and customizable vibration alerts through a wearable band [1]. The system uses speech recognition, natural language processing, and audio classification while prioritizing on-device processing for privacy and low latency. Development involved challenges integrating these components and debugging code as new features were added [2]. Two experiments tested the system under realistic conditions. Sound detection reached 90% accuracy in quiet conditions but decreased to 60% with conversation noise.Transcription achieved a 6% word error rate for clear speech but reached 35% when fast, accented speech and background noise were combined. These results highlight areas for continued improvement while demonstrating hearassist’s potential to combine multiple accessibility functions into one system for everyday communication and environmental awareness.

KEYWORDS

Assistance, Hearing, Accessibility, Convenience


Flow A User-Friendly Program to Help Gamers Discover, Compare, And Save Games Using HTML, CSS, JavaScript, Firebase, And Game Databases

Haoyi Wang, Armando Contreras, 1St Margaret’s Episcopal School, 31641 La Novia Avenue, San Juan Capistrano, CA 92675, 2University of California, Irvine, Irvine, CA 92697

ABSTRACT

Video game stores have become really crowded since platforms like Steam let people find, buy, and talk about games all in one place (Valve Corporation, n.d.). My project, Uptrend, tries to make finding new games easier and less confusing. Uptrend shows popular games, trends, data, ratings, comments, and also has a watchlist. I built the website using HTML, CSS, JavaScript, Firebase Authentication, Firestore, and game data links (Google, 2026a; Google, 2026b) [13]. While making it, I had problems with spacing, old game visuals, and Firebase permissions. I fixed these by cleaning up the layout, updating the game data, and changing how the login and watchlist syncing worked. When I tested Uptrend, users were able to find game pages faster than with a normal search, and the watchlist saved games quickly before syncing them online [1]. Overall, Uptrend makes finding and keeping track of games faster, simpler, and easier to understand.

KEYWORDS

Video games, Tracking, Discovery, Marketplace, HTML, CSS, JavaScript, Friebase, Data


Flow State: Design and Evaluation of a Mobile Application for Tennis Player Matchmaking and Session Scheduling

Elvin Qihuo Liao1, Bobby Nguyen2, 1Sage Hill High School, Newport Coast Drive, Newport Beach, CA, 2University of California, Riverside, 900 University Ave, Riverside, CA 92521

ABSTRACT

This paper presents Flow State, a mobile application designed to help tennis players find compatible partners and organize sessions. The project addresses the difficulty beginners and independent players experience when searching for opponents with suitable skill levels, locations, schedules, and playing preferences. The application combines three major components: user authentication and profile management, a compatibility-based matchmaking system, and session booking and scheduling. Flutter and Dart provide the application interface and logic, while Firebase Authentication, Cloud Firestore, and Firebase Storage manage accounts and data [4]. Match recommendations use weighted profile, session, history, and recent-activity information, including filtering for inactive profiles. Booking features check capacity, duplicate reservations, and schedule conflicts. Experiments using simulated player profiles and scheduling cases tested recommendation alignment and booking reliability. Results indicated that the system could produce personalized matches and reject invalid bookings. Future improvements include larger real-use testing, skill verification, calendar integration, and stronger privacy and safety features.

KEYWORDS

Sports Matchmaking, Tennis, Compatibility-Based Recommendation, Mobile Application Development


A Network-Based Epidemic Management Simulation to Model Public Health Intervention Trade-offs Using Unity and C#

Jinxi Sha1, Moddwyn Andaya2, 1Sage Hill School, 20402 Newport Coast Drive, Newport Coast, CA 92657, 2California State University, Sacramento, 6000 Jed Smith Dr, Sacramento, CA 95819

ABSTRACT

Virus VS Vaccine is a Unity-based strategy simulation in which a player manages a spreading virus across a network of connected locations by allocating a limited weekly budget of Efficiency Points toward vaccination and shutdown interventions. The project responds to the ongoing real-world need for tools that make networked, resource- constrained public health trade-offs tangible rather than abstract. We describe the three linked systems that make this possible: a location-and-connection network, a shutdown intervention state machine, and an Efficiency Point economy, and walk through a representative code sample from each. An experiment comparing network-severing interventions against vaccination found vaccination consistently more effective at limiting total infections, a result consistent with published findings on the limited effectiveness of travel restrictions alone. We compare our approach against three existing methodologies, three compartmental modeling, cellular-automata simulation, and branching-scenario games, and outline the calibration and state-management improvements that further development would prioritize.

KEYWORDS

Epidemic simulation, Game-based learning, Compartmental modeling, Public health intervention


Common Matters: An Adaptive Mobile Application for Teaching Everyday Life Skills to Children using a Large Language Model with Multi-Modal Learning-Style Personalization

Lanqi Li1, Rodrigo Onate2, 1Eleanor Roosevelt High School, 7447 Scholar way, Eastvale, CA 92880, 2California State University, Fullerton, 800 N State College Blvd, Fullerton, CA 92831

ABSTRACT

Everyday life skills are rarely taught with the same individualisation that academic subjects increasingly receive, and existing children’s apps either present identical content to every user or expose general-purpose chatbots without age-appropriate guardrails. This paper presents Common Matters, an adaptive Flutter application that uses GPT-5.2 to generate ten-paragraph life-skills lessons whose vocabulary, narrative form, and assessment shape are all conditioned on the child’s saved reading level and one of six learning styles, with illustrations served from a SHA-1 keyed Firebase Storage cache and fill-in-the-blank answers graded by the model itself. Two evaluations on the live system found that generated stories were valid and on topic in over 94% of 300 trials, and that the LLM grader agreed with human raters above 91% on every legitimately correct answer variation while strict matching fell below 20%. Common Matters offers a low-cost, inclusive alternative to static educational apps.

KEYWORDS

Large Language Models, Educational Technology, Adaptive Learning, Learning Styles, Flutter, Firebase, Children’ s Education, Personalized Content Generation, AI-Graded Assessment, Mobile Learning


XXX InstructFace-Edit-XL: A Quality-AnnotatedFacial Editing Dataset, Curated with Metrics,VQA, and Human Review

Viktoriia Vovchenko , Vincenzo Barberi, Sergej Schultenkämper, Frederik ,Hochschule Bielefeld - Bielefeld University of Applied Sciences and Arts (HSBI),Germany

ABSTRACT

Instruction-guided image editing models have made photorealistic facial manipulation accessi-ble through natural language prompts, substantially lowering the barrier to misuse. These systems facilitateattacks such as identity theft, phishing, misinformation, and social engineering, while also challenging ex-isting forensic methods that were not designed for prompt-driven edits. XXX InstructFace-Edit (XIFE) isa dataset of 36,000 edited facial image pairs at 1024 × 1024 resolution generated by four state-of-the-artinstruction-guided image editing models (FLUX.1-Kontext-dev, FLUX.2-dev, Qwen-Image-Edit-2511, andStep1X-Edit-v1.2) across six semantically distinct edit types. It was curated using a multi-stage filteringpipeline that integrates structural, semantic, and multimodal large language model-based quality checks,with thresholds derived empirically from a human-annotated subset of 1,500 images. However, the paceat which new editing models are released routinely outstrips the construction cycle of any single dataset,so we extend XIFE with a fifth model, HunyuanImage 3.0, released shortly after the original dataset wascompleted and since then a leader on public image-editing leaderboards. To curate this extension, we in-troduce a new evaluation pipeline that combines human review with visual question answering (VQA),set it against the original VIEScore-based pipeline, and set out the design trade-offs between them. Theextension generates 17,868 edited pairs, of which 3,064 pass curation and 14,804 are retained under anexplicit unacceptable label. The collection now totals 53,868 facial image pairs across five editing modelsand six edit types, 21,832 of them annotated as high quality. The full dataset, including all quality labels,is publicly available for training and stress-testing detectors for instruction-guided facial manipulation.

KEYWORDS

Dataset, Image forensics, Instruction-guided image editing, Visual question answering


Next-Generation Networks for Image Recognition

Jan Flusser and Vaclav Kosik, Czech Academy of Sciences, Institute of Information Theory and Automation, PodVodárenskou věží 4, Prague 8, 182 08, Czech Republic

ABSTRACT

We propose novel hybrid network architectures for imagerecognition. Hybrid networks work not only on pixel level as traditionalCNN do but in parallel they work with higher-level representation, invari-ant to intra-class variabilities. Outcomes of both parts of the network arefused together to obtain robust final classification. The main advantageof this approach is a significant reduction of the training set augmen-tation and thus saving a substantial portion of the training time whilemaintaining high recognition rate.

KEYWORDS

Image recognition · Hybrid networks · Invariant represen-tation · Fusion.

menu
Reach Us

emailaiaa@aiaa2026.org


emailaiaaconference@yahoo.com
close